International Conference on Advanced Intelligent Systems for Sustainable Development: Volume 3 - Advanced Intelligent Systems on Agriculture and Health (Lecture Notes in Networks and Systems, 713) 3031352475, 9783031352478
This book describes the potential contributions of emerging technologies in different fields as well as the opportunitie
130 80 83MB
English Pages 886 [882] Year 2023
Foreword
Preface
Organization
Acknowledgement
Contents
Breast Cancer Progression Prediction for Care Treatment Efficiency and Intelligent Adaptation, Based on AI Algorithm Classification
1 Introduction
2 Related Works
3 Proposed Approach
3.1 Assessment of Response to Treatment
3.2 Radiomic or Computational Medical
4 Result and Discussion
5 Conclusion and Perspectives
References
Study of the Germination of Wild and Cultivated Blackberries of the Northern Region of Morocco
1 Introduction
2 Methods
2.1 Seed Materials
2.2 Seed Preparation
2.3 Preparation of Hormones and Sterilization of Equipment
2.4 Germination Experiments
2.5 Germination Treatment
2.6 Data Treatment
3 Results
3.1 Imbibition Test
3.2 Gibberellic Acid Treatment
3.3 Treatment of Rubus Seeds with Sulfuric Acid (H2SO4) Combined with Boiling Water
4 Conclusion
References
The Impact of Covid 19 on Recommendation Platforms
1 Introduction
2 Background
2.1 Recommender Systems
2.2 Cold Start Users
2.3 Grey Shep Users
3 Recommender Systems and Covid 19
4 Cold Start and Grey Shep Users
5 Proposed Solution Based on Adaptive Recommendation Algorithms
6 Conclusion
References
Classifiers-Based Personality Disorders Detection
1 Introduction
2 Related Work
3 Our Approach
3.1 Approach Overview
3.2 Unit Classification Modules
4 Prototype, Experiments and Results
4.1 Datasets Description
4.2 Experimental Evaluation
4.3 Clustering and Feature Selection
4.4 Clusters Interpretation
5 Conclusion
References
How Health Information Technology Improved Patient Care and Treatment During the COVID-19 Pandemic: A Comparison Between International Case Studies and the Moroccan Context
1 Introduction
2 Background
3 Research Method
3.1 The Role and Impact of HIT on Care Delivery During the Present COVID-19 Crisis and the Breadth of HIT in the Health Care Sector During the Pandemic from an International Perspective
3.2 How Developing Countries, Such as Morocco, Have Responded to Covid 19 by Over- Hauling and Modernizing Their Hospital Information Systems at the Operational, Tac- Tical, and Strategic Levels.
3.3 Morocco’s Accelerated Path to Digital Health Transformation
4 Conclusion
References
Dates Detection System for Automatic Harvesting Using Deep Learning
1 Introduction
2 The Problem Formulation
2.1 CNN Deep Learning
2.2 Object Detection
2.3 Yolo
3 The Proposed Solution
3.1 Dataset
3.2 Network Parameterization and Training Details
3.3 Result and Discussion
4 Conclusions
References
Machine Learning for Diabetes Prediction
1 Introduction
2 Research Methodology
2.1 Define the Research Problem
2.2 Setting Research Questions and Objectives
2.3 Criteria Identification
2.4 Pertinent Literature Extraction
2.5 Literature Synthesis and Discussion
3 Conceptual Framework for Diabete Prediction
4 Conclusion
References
Technology as an Answer to the Trust Crisis in Mental Health Services - Digitization Serving Mental Health Care Systems
1 Introduction
2 Technology Serving Mental Health Care Systems
2.1 Revolutionary Potential
2.2 Legal Concerns
2.3 Trust in Mental Health Technology
3 Practical Examples of Technology Use for Mental Health
4 Discussion
References
Artificial Intelligence at the Service of Precision Medicine
1 Introduction
2 Research Methodology
2.1 Literature Review Planning Protocol
2.2 Pertinent Literature Extraction
3 Results of the Literature Review
3.1 Machine Learning Methods Analysis
3.2 Random Forests
3.3 Support Vector Machines (SVM)
3.4 Decision Tree
3.5 K-means
3.6 Deep Learning (DL)
3.7 Artificial Neural Network (ANN)
3.8 Convolutional Neural Network (CNN)
3.9 Recurrent Neural Network (RNN)
4 Conclusion
References
Case-Based Reasoning Approach, Integrating Deep Learning for Patient Diagnosis Combined X-Ray with Symptoms
1 Introduction
2 Overview of Diagnostics Systems, Knowledge-Based and Non-knowledge-Based Reasoning in Healthcare
3 Introduction to CBR, Deep Learning Systems
3.1 Case Based Reasoning
3.2 Deep Learning
4 The Proposed Architecture for the Improvement of the Diagnostic Support System
5 Methodology
6 Conclusion
References
AGRI-PREDI Prediction System of Climate Change Based on Machine Learning for Precision Agriculture in Mediterranean Region
1 Introduction
2 Area of Application
2.1 Geographical Application: The Mediterranean Region
2.2 Type of Crop in Application: Olives
3 Building a New Logical Rule Based on Climate Change Data
4 Building a New Logical Rule Based on Climate Change Data
5 Results and Discussion
6 Conclusion
References
Estimated Daily Reference Evapotranspiration Using Machine Learning and Deep Learning Based on Various Combinations of Meteorological Data
1 Introduction
2 Materials and Methods
2.1 Dataset and Data Combinations
2.2 Data Cleaning
2.3 Computation of Reference Evapotranspiration ET0
2.4 Long Short-Term Memory Neural Network
2.5 Evaluation Metrics
3 Results and Discussion
3.1 Scenario A
3.2 Scenario B
3.3 Model Validation
4 Conclusion
References
Deep Learning Approach for Brain Tumor Classification Implemented in Raspberry Pi
1 Introduction
2 Materials and Methods
2.1 Hardware
2.2 Software
3 Dataset
4 Results and Discussions
5 Conclusion
References
Machine Learning for Medical Image Analysis: A Survey
1 Introduction
2 Medical Image Analysis
2.1 Image Compression
2.2 Image Classification
2.3 Image Segmentation
2.4 Image Detection
3 Machine Learning Techniques
3.1 Supervised Learning
3.2 Unsupervised Learning
3.3 Semi- Supervised Learning
3.4 Reinforcement Learning
3.5 Deep Learning
4 Machine Learning Approaches for Medical Image Processing
5 Discussion
5.1 Image Compression
5.2 Image Classification
5.3 Image Segmentation
5.4 Image Detection
6 Conclusion
References
Simulation Study on Validating an Efficient Technique for Early Breast Cancer Detection Using Thermographic Micro-biosensors
1 Introduction
2 Description of the Proposed Technique
3 Numerical Breast Tumors Simulations
3.1 The Physical Properties of the Breast
3.2 COMSOL-Based Breast Tumors Simulations
4 Experimental Test
4.1 Breast Phantom Construction
4.2 Micro-biosensors Implementation
5 Conclusion
References
Argania Forest Change Detection from Sentinel-2 Satellite Images Using U-Net Architectures
1 Introduction
2 Material and Methods
2.1 Study Area: Admine Forest
2.2 Dataset
2.3 Deep Learning Models
2.4 Hyperparameters
3 Results and Discussion
3.1 Classification
3.2 Change Detection
4 Conclusions
References
Development of Environmental Performance Measurement Model for Public Hospitals
1 Introduction
2 Theoretical Framework
2.1 A General Overview of Healthcare Setting
2.2 Performance Measurement System
2.3 Environmental Performance Measurement in Healthcare Organizations
3 Methodology
3.1 Balanced Scorecard (BSC)
3.2 Fuzzy Analytic Hierarchy Process (FAHP)
4 Case Study
5 Conclusion
References
The Nitrogen Effects on Growth and Development of Morphological Parameters of “Argania Spinosa L. skeel” Tree Seedlings
1 Introduction
2 Material and Methods
2.1 Study Area of “Argania spinosa L. skeel” Seeds Materials
2.2 Experimental Site
2.3 Plant Material
2.4 Substratum
2.5 Media Analysis
2.6 Germination and Nutrient Solution Tested
2.7 Plant Development Measurements and Monitoring
2.8 Experimental Apparatus
2.9 Data Analysis
3 Results
3.1 Media Analysis
3.2 Plant Development
4 Discussion
5 Conclusion
References
Cost-Effective Manufacturing Operations During and After the COVID‑19 Pandemic
1 Introduction
2 Review Methodology
3 Smart Manufacturing and Epidemic/pandemic
3.1 Manufacturing Operations During Epidemic/pandemic
3.2 Manufacturing Goals and the Importance of Coordination
3.3 Epidemic/pandemic Control Stages
3.4 AI-Assisted Solutions for Smart Manufacturing
4 Analysis of Results
5 Conclusion
References
Hypertension Risk Prevention: Towards a Predictive Model Based on Data Mining
1 Introduction
2 State of the Art
3 Approach Taken
3.1 Overview of the Extraction Process
3.2 Data Collection
3.3 Statistical Analysis
4 Conclusion and Perspectives
References
A Lightweight CNN Model for Tomato Crop Disease Detection
1 Introduction
2 Data
3 Results and Analysis
4 Conclusion
References
Discrete Event Simulation for the Evaluation of Emergency Department Layout
1 Introduction
2 Literature Review
3 Problem Description
4 Problem Modeling and Resolution
5 Conclusion
References
Word Sense Disambiguation in the Biomedical Domain: Short Literature Review
1 Introduction
2 Problem Statement
3 Methods
3.1 Knowledge Based Methods
3.2 Supervised Learning Methods
3.3 Unsupervised Learning Methods
3.4 WSD Using Bilingual Corpora
4 Available Datasets
5 Discussion
6 Conclusion and Future Work
References
Environmental Impact Assessment of Agricultural Practices Using the Life Cycle Assessment Method: A Review
1 Introduction
2 Environmental Analysis of Agricultural Systems
3 Beet (Beta Vulgaris)
3.1 Beet Farming Phase
4 Materials and Methods
4.1 Definition of Purpose and Scope
4.2 Life Cycle Inventory
4.3 Life Cycle Impact Assessment
4.4 Interpretation
5 Uncertainty Analysis
6 Results and Discussion
7 Conclusion
References
Weakly Supervised Patchwise CT Liver Segmentation
1 Introduction
2 Related Works
2.1 Traditional Segmentation Methods
2.2 Fully Supervised Segmentation Methods
2.3 Segmentation Methods that are Weakly Supervised
3 Method
3.1 Patch-Based Training
3.2 Reiterative Learning Framework
3.3 BoxUNet Architecture
3.4 Postprocessing
4 Evaluation and Results
4.1 Dataset
4.2 Conditional Random Field for Postprocessing
5 Conclusions
References
Smart Agriculture Using an Enhanced Selection for MPRs
1 Introduction
2 Related Work
3 Proposed Mechanism
3.1 Brief Examination of OLSR
3.2 Proposed Mechanism
3.3 Proposed Algorithm
4 Results and Analysis
4.1 Simulation Mobility Model
4.2 Comparison and Discussion
5 Conclusion
References
Prevalence of Anxiety and Depressive Disorders and Cognitive Affective Risk Factors: Case of Officials of the Rabat-Sale-Kenitra Region
1 Introduction
1.1 Purpose of the Study
1.2 Study Population
2 Method
3 Results
3.1 Socio-demographic Profile of the Subjects
3.2 Prevalence of Anxiety-Depressive Disorders
3.3 Bivariate Analyses
4 Discussion
5 Conclusion
References
Predicting Drug Compounds Effectiveness Based on Chemical Properties and Bioactivity Data
1 Introduction
2 Related Work
3 Proposed Approach and Methodology
3.1 Methodology
3.2 Molecular Fingerprints
3.3 Exploratory Analysis
4 Prediction Results and Discussion
5 Conclusion
References
A Review of the Transportation Routing Problem During the COVID-19 Pandemic
1 Introduction
2 Research and Classification Methodology
2.1 Research Methodology
2.2 Classification by the year and Publication Medium
2.3 Classification by Field of Activity
3 Vehicles Routing Problem in COVID-19 Pandemic Context
3.1 Classification According to Business Constraints
3.2 Classification According to the Methods of Resolution
4 Conclusion
References
The Effect of a Physical Activity Program on Improving Sports Performance in Obese Adolescents Based on a Statistical Analysis of Cardiovascular Physiological Signals
1 Introduction
2 Approach Taken
2.1 Subjects and Methods
2.2 Results
3 Discussion
4 Conclusion
References
HER-Omics, a Model of Transcriptomics Data Integration in EHRs
1 Introduction
2 Methods
2.1 Tool Construction
2.2 Performed Analysis
3 Results
4 Discussion
References
Artificial Intelligence for Stroke Prediction
1 Introduction
2 Methods and Materials
2.1 Dataset
2.2 Relation Between Variables
3 Results and Discussion
3.1 Preprocessing Steps
3.2 Results
4 Conclusion
References
The Role of Uncertainty Avoidance, Trust, and Land Tenure in Predicting the Adoption of Green IoT Irrigation Systems in Morocco: An Improved Model
1 Introduction
2 Context
3 Theoretical Framework
3.1 TAM’s Extension by Trust and Hofstede’s Cultural Dimension: Focus on Uncertainty Avoidance and Collectivism
3.2 The Influence of Land Tenure in IT Adoption in the Agricultural Sector
4 Improved Research Model and Hypothesis
5 Methodology Adopted
6 Conclusion
References
The Complete Mitochondrial Genome of Agapornis roseicollis
1 Introduction
2 Methods
3 Results
4 Discussion
5 Conclusion and Perspectives
References:
Cross-Talk Between Intramolecular and Intermolecular Amino Acid Networks Orchestrates the Assembly of the Cholera Toxin B Pentamer via the Residue His94
1 Introduction
2 Method
2.1 Fold-X
2.2 Gemini
2.3 Spectral-Pro
3 Results
4 Conclusion
Appendix
References
Identification of Responsibilities for the Use of Devices Incorporating Artificial Intelligence in Health: Systemic Review
1 Introduction
2 Health AI Applications
3 Materials and Methods
4 Résultats
4.1 Data Classification
4.2 Issues Related to AI
4.3 Stakeholders
4.4 Responsibilities for the Use of Intelligent Devices (ID)
4.5 Distribution of Stakeholder Responsibilities
4.6 Conceptual Data Model
5 Conclusion
References
Artificial Intelligence Recognition of Human Body Actions for Bionic Applications (New Combination of Indicator Parameters)
1 Introduction
2 System Design and Equipments
3 Proposed Methodology
3.1 Data Capture
3.2 Signal Processing and Filtration
3.3 Feature Extraction
3.4 Data Classification
4 Experiments and Result
5 Conclusion and Further Research
References
Parkinson’s Disease Recognition from Speech Signal Using Discrete Wavelet Transform, Delta, Delta-Delta, and K-Nearest Neighbor
1 Introduction
2 Theoretical Context
2.1 Sample Collection
2.2 Feature Extraction
2.3 Classification
3 Methodology and Results
4 Discussion
5 Conclusions
References
Triple Negative Breast Cancer and Non-Triple Negative Breast Cancer Recurrence Prediction Using Boosting Models
1 Introduction
2 Related Works
3 Proposed Approach
3.1 Dataset and Pre-processing
3.2 Work-Flow
4 Experiment Results and Discussion
4.1 Evaluation Metrics
4.2 Results
4.3 Discussion
5 Conclusion
References
A Word-Based Moroccan Sign Language Dataset for Emergency Situations: Use Case COVID-19
1 Introduction
2 Related Works
3 The Moroccan Sign Language Database
3.1 Value of the Data
3.2 Data Description
4 Experimental Design
5 Conclusion
References
The Use of ISFET for the Measurement of Phosphorus in Moroccan Soils
1 Introduction
2 Material and Method
2.1 Soil Sample Preparation
2.2 Soil Extract Preparation
2.3 ISFET Used
2.4 Calibration System
3 Results and Discuss
3.1 Calibration Curves
3.2 Measurement in Known Concentrations of H2PO4-
3.3 Extract Soil Measurement
4 Conclusion
References
Exploring the Performance of Traditional Word Embeddings and Contextual Embedding Models for COVID-19 Sentiment Analysis
1 Introduction
2 Related Works
3 The Proposed Approach
4 Results
4.1 Preprocessing Covid-19 Dataset
4.2 Exploratory Analysis
4.3 Result and Discussion
5 Conclusion
References
Mycoflora of Dormant Crocus Sativus Corms in Morocco
1 Introduction
2 Material and Methods
2.1 Sampling
2.2 Isolation
2.3 Morphological Identification of Isolated Fungal Species
3 Results
4 Discussion
5 Conclusion
References
Optimized DL-Based Model for Hypertrophic Cardiomyopathy CMR Images
1 Introduction
2 Methodology
2.1 Dataset Description
2.2 The Proposed CNN Architecture
2.3 Data Augmentation Techniques and Optimizers
3 Results and Discussion
4 Conclusion
References
Low-Cost Water Conductivity Sensor Based on a Parallel Plate Capacitor for Precision Agriculture
1 Introduction
2 Related Work
3 System Overview
4 Principle of Operation of Our Capacity-Based Salinity Sensor
4.1 Physical Description and Mathematical Model
4.2 Combining Capacitors
5 Salinity Sensor Based on Parallel Plate Capacitor
5.1 Physical Description
5.2 Electrical Circuit of Salinity Sensor
6 Measurements Results
6.1 Measurements of Alt Concentration in Water
6.2 Network Coverage
7 Conclusion and Future Work
References
Literature Review of Deep Learning for Tuberculosis Based on Chest Imaging
1 Introduction
2 Research Methodology
2.1 Research Questions
2.2 Search Strategy
3 Results and Discussion
4 Conclusion
References
Inventory of Corticolous Lichens and Argan Wood Fungi
1 Introduction
2 Materials and Methods
3 Results
4 Discussion
5 Conclusion
References
Detection of Alpha-Cypermethrin in Food Using a Sensor Array Coupled to Machine Learning Algorithms: The Case of the Wild Edible Swiss Chard
1 Introduction
2 Material and Methods
2.1 Preparation of Samples
2.2 Experimental Design and Working Principal
2.3 Statistical Analysis
3 Results and Discussions
3.1 The Sensor Network Responses to VOC Emissions from Wild Swiss Chard
3.2 PCA Result
3.3 KNN Result
4 Conclusion
References
Effect of Some Derivatives of Pyridazin-3 (2h) – Ones on the in Vitro and in Situ Development of Different Pathogenic Fungi on Citrus Fruits
1 Introduction
2 Materials and Methods
2.1 The Pathogens Tested
2.2 The Molecules Tested and Preparation of the Culture Medium
2.3 Effect of the Five Molecules on Mycelial Growth
2.4 Effect of Pyridazine Molecules on Sporulation
2.5 Development of an In-Situ Screening Test for the Effect of Pyridazine Compounds on Phytophthora Spp. on Chopsticks of Rought Lemon Rootstock Citrus
2.6 Statistical Analysis
3 Results
4 Discussion
5 Conclusion
References
Recovery of Organic Waste by Biogas Production-Mathematical Modeling of Anaerobic Digestion: A Short Literature Review
1 Introduction
2 Literature Review
2.1 Mechanistic Models
2.2 Data-Driven Models
3 Discussion and Perspectives
4 Conclusion
References
Automation of Agriculture Using Artificial Intelligence: Towards a Sustainable Agriculture in Morocco
1 Introduction
2 Internet of Things and Artificial Intelligent in Agriculture
3 Soil and Irrigation Management
4 Pest and Weed Control
5 Conclusion
References
Review of Weed Detection Methods Based on Machine Learning Models
1 Introduction
2 Related Work
2.1 The Weeds Detection
2.2 ML and DL Techniques
2.3 Deep Learning Methods for Detecting and Identifying Weeds
3 Analysis and Discussions
3.1 Various Weed Detection Tasks
3.2 Several Complicated Factors Affecting Weed Detection
4 Conclusion
References
Analysis of Alcoholic EEG Signals Based on Discrete to Continuous Algorithm
1 Introduction
2 Method
2.1 Discrete to Continuous Algorithm
2.2 Dynamic Time Warping Algorithm
2.3 Data Description
3 Experimental Results
4 Discussion
5 Conclusion
References
Growth Promoting of Tomato Plants by Incorporation of Trichoderma asperellum Enriched Liquid Product via Foliar Spray and the Irrigation System
1 Introduction
2 Materials and Methods
3 Results and Discussion
References
Effect of Trichoderma Asperellum on the Development of Strawberry Plants and Biocontrol of Anthracnose Disease Caused by Colletotrichum Gloeosporioides
1 Introduction
2 Material and Methods
2.1 Fungal Material
2.2 Plant Material
2.3 Preparation of the Inoculum
2.4 Inoculation and Treatments
2.5 Scoring of Results
2.6 Statistical Analysis
3 Results and Discussion
4 Conclusion
References
Diversity of Arbuscular Mycorrhizal Fungi in the Rhizosphere of Four Citrus Genotypes
1 Introduction
2 Materials and Methods
2.1 Surveys and Sampling
2.2 Root Staining
2.3 Evaluation of the Mycorrhization Rate
2.4 Spores Extraction
2.5 Statistical Analysis
3 Results and Discussion
3.1 Characterisation of the Mycorrhizal Load of Cultivated Citrus
3.2 Spores Diversity of Arbuscular Mycorrhizal Fungi (AMF)
4 Conclusion
References
Contribution to the Development of a Technological Platform for Analysis in Precision Agriculture for the Biovigilance of Cryptogamic Diseases in Strawberry (Fragaria × ananassa)
1 Introduction
2 Materials and Methods
2.1 Location
2.2 Data Management and Processing
3 Result and Discussion
4 Conclusion
References
Diversity of the Bryoflora of the Tazekka National Park, North-East Morocco
1 Introduction
2 Material and Methods
2.1 Presentation of the Study Area
2.2 Methodology
2.3 Data Analysis
3 Results
4 Discussion
5 Conclusion
Appendix
References
A New Model-Based Approach for Migrating Health 2.0 to Health 3.0 Applications
1 Introduction
2 Context
2.1 A. E-health and Semantic Technologies
2.2 Standards Based on ADM
3 The Proposed Approach
3.1 Reengineering Phase
3.2 Restructuring Phase
3.3 Forward Engineering Phase
4 Case Study and Discussion
5 Related Work
6 Conclusion
References
Effect of the Impact of Environmental Factors on the Diet of Horse Mackerel (Trachurus trachurus) from the Coasts of the Moroccan
1 Introduction
2 Materials and Methods
2.1 Presentation of the Study Area: Cap Spartel- Casablanca
2.2 Sampling
2.3 Examination of Stomach Content
2.4 Data Analyses
3 Results and Discussion
3.1 Specimens Size
3.2 Dietary Intensity and Feeding Behavior
3.3 Seasonal Variation of Dietary
3.4 Concentration of Chlorophyll a and Oxygen as a Function of Pray
4 Conclusion
References
Study of the Risks Associated with the Use of Pesticides on Soils and Groundwater in the Loukkos Perimeter (Morocco)
1 Introduction
2 Materials and Methods
2.1 Presentation of the Study Area: Provinces of Larache
2.2 Sampling
2.3 Extraction of Pesticides According to the 15662 QuEChERS Method
3 Results and Discussion
3.1 The First Modification
3.2 The Second Modification
3.3 Result of the Improvement of the QuECHeRS Extraction Method
4 Conclusion
References
Diversity of the Bryoflora of a Cedar Forest in the Central High Rif, Northern Morocco
1 Introduction
2 Material and Methods
2.1 Presentation of the Study Area
2.2 Collecting Plant Material
2.3 Data Analysis
3 Results
4 Discussion
5 Conclusion
Appendix
References
Toward a Smart Agriculture Using Deep Learning for Plant Disease Detection: Challenges and Opportunities
1 Introduction
2 Background: Deep Learning and IoT for Smart Agriculture/farmer
3 A Case Study: Disease Detection State of the Art
4 Applications of Deep Learning in Agriculture
5 Conclusion
References
Study of the Effect of Plastic Film in Tunnels on the Yield and Quality of Strawberry Plants
1 Introduction
2 Material and Methods
2.1 Plant Material: Variety Sabrina
2.2 The Plastic Used
2.3 Analyzes of Some Physico-Chemical Parameters:
2.4 Analysis of Some Fruit Quality Parameters
3 Results and Discussion
4 Conclusion and Prospects
References
Effect of Endomycorrhizae on the Pathogenicity of Rhizoctonia solani and the Growth of Olive Plants
1 Introduction
2 Materials and Methods
2.1 Preparation of the Mycorrhizal Inoculum
2.2 Preparation of Fungal Inoculum
2.3 Inoculation of Olive Plants
2.4 Evaluation of Mycorrhization Parameters
3 Results
4 Discussion
5 Conclusion
References
Detection of Alzheimer's Disease Using a Convolutional Neural Network
1 Introduction
2 Image Database
3 Method
3.1 Problem
3.2 Initial Architecture with no Hidden Layers
3.3 Adding Fully Connected Hidden Layers
3.4 Convolutional Neural Network
4 Results and Discussion
5 Conclusion
References
Distribution of Resellers and Modeling of the Future of the Majority Pesticides for Agricultural Use in Morocco
1 Introduction
2 Material and Methods
3 Results and Discussion
3.1 Type of Pesticides Marketed in Laouamra
3.2 Classification of the Dominant Phytosanitary Products in Laouamra
4 Conclusion and Prospects
References
A Proposed Architecture Based on Deep Learning and Optimization Techniques for Medical Diagnostic Imaging
1 Introduction
2 Related Works
3 Proposed Architecture
3.1 Pre-processing
3.2 Feature Extraction
3.3 Feature Selection: Chaotic-MRFO
3.4 Classification
4 Experimental Evaluation and Result
4.1 Data Preparation and Augmentation
4.2 Performance Measures
5 Conclusion and Future Work
References
A Hierarchical Machine Learning Algorithm for Epileptic Seizure Prediction
1 Introduction
2 Related Works
3 Machine Learning for Medical Sudden Emergency Prediction
3.1 Algorithms
3.2 Datasets
4 Hierarchical Algorithm for Epilepsy Prediction
4.1 Tachycardia and Epilipsy Data and Phases
4.2 Explaining the Process of the Hierarchical Algorithm
5 Simulation and Results
5.1 Data and Simulation
5.2 Results and Discussion
6 Conclusion
References
Epileptic Seizure Prediction Using Artificial Intelligence Methods
1 Introduction
2 Data Set Description
3 Artificial Intelligence Methods
3.1 K-Nearest Neighbors
3.2 Logistic Regression
3.3 Support Vector Machine (SVM)
3.4 Gaussian Naive Bayes
3.5 Neural Network
References
New Approach of 3D Protein Structure Superimposition: Case Study of “SARS-COV-2” and “SARS-COV”
1 Introduction
2 Algorithms
3 Results
3.1 ACE2 Alignment
3.2 Spike Alignment
3.3 Total Complex Alignment (ACE2+ SARS and ACE2+ SARS-COV2)
3.4 Our Proposed Solution
4 Discussion
5 Conclusion
References
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application in Biomedicine
1 Introduction
2 Experimental
2.1 Synthesis of Materials Based on Gadolinium and Cerium SrMP2O7 (M = Gd or Ce)
3 Results and Interpretation
3.1 X-Ray Diffraction
3.2 Optical properties of Pyrophosphates SrMP2O7 with M = Gd and Ce Transmittance Spectra
3.3 Infrared Spectroscopy
4 Discussion
5 Conclusion
References
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization and Advanced Structure Study Application: Improve Mechanical Strength of Implant and the Osteointegration Process
1 Introduction
1.1 Mechanical Stability (Primary Stability)
1.2 Secondary Stability (Biological Stability)
2 Experimental
2.1 Synthesis of Niobium Triphosphates NbP3O10
2.2 Characterization of Niobium Triphosphates NbP3O10
3 System Design, Mesh and Finite Element Modeling (FEM)
References
Extraction and Characterization of New Cellulosic Fibers from Moroccan Mallow Stem and Comparison with Other Naturel Fibers
1 Introduction
2 Experimental Processes
3 X-Rray Fluorescence
3.1 X-Ray Fluorescence (XRF) Analysis
3.2 X-Ray Diffraction (XRD) Analysis
3.3 Fourier Transformed Infra-red (FTIR) Spectral Analysis
4 Conclusion
References
Author Index
Recommend Papers
File loading please wait...
Citation preview
Lecture Notes in Networks and Systems 713
Janusz Kacprzyk Mostafa Ezziyyani Valentina Emilia Balas Editors
International Conference on Advanced Intelligent Systems for Sustainable Development Volume 3 - Advanced Intelligent Systems on Agriculture and Health
Lecture Notes in Networks and Systems
713
Series Editor Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland
Advisory Editors Fernando Gomide, Department of Computer Engineering and Automation—DCA, School of Electrical and Computer Engineering—FEEC, University of Campinas—UNICAMP, São Paulo, Brazil Okyay Kaynak, Department of Electrical and Electronic Engineering, Bogazici University, Istanbul, Türkiye Derong Liu, Department of Electrical and Computer Engineering, University of Illinois at Chicago, Chicago, USA Institute of Automation, Chinese Academy of Sciences, Beijing, China Witold Pedrycz, Department of Electrical and Computer Engineering, University of Alberta, Alberta, Canada Systems Research Institute, Polish Academy of Sciences, Warsaw, Poland Marios M. Polycarpou, Department of Electrical and Computer Engineering, KIOS Research Center for Intelligent Systems and Networks, University of Cyprus, Nicosia, Cyprus Imre J. Rudas, Óbuda University, Budapest, Hungary Jun Wang, Department of Computer Science, City University of Hong Kong, Kowloon, Hong Kong
The series “Lecture Notes in Networks and Systems” publishes the latest developments in Networks and Systems—quickly, informally and with high quality. Original research reported in proceedings and post-proceedings represents the core of LNNS. Volumes published in LNNS embrace all aspects and subfields of, as well as new challenges in, Networks and Systems. The series contains proceedings and edited volumes in systems and networks, spanning the areas of Cyber-Physical Systems, Autonomous Systems, Sensor Networks, Control Systems, Energy Systems, Automotive Systems, Biological Systems, Vehicular Networking and Connected Vehicles, Aerospace Systems, Automation, Manufacturing, Smart Grids, Nonlinear Systems, Power Systems, Robotics, Social Systems, Economic Systems and other. Of particular value to both the contributors and the readership are the short publication timeframe and the world-wide distribution and exposure which enable both a wide and rapid dissemination of research output. The series covers the theory, applications, and perspectives on the state of the art and future developments relevant to systems and networks, decision making, control, complex processes and related areas, as embedded in the fields of interdisciplinary and applied sciences, engineering, computer science, physics, economics, social, and life sciences, as well as the paradigms and methodologies behind them. Indexed by SCOPUS, INSPEC, WTI Frankfurt eG, zbMATH, SCImago. All books published in the series are submitted for consideration in Web of Science. For proposals from Asia please contact Aninda Bose ([email protected]).
Janusz Kacprzyk · Mostafa Ezziyyani · Valentina Emilia Balas Editors
International Conference on Advanced Intelligent Systems for Sustainable Development Volume 3 - Advanced Intelligent Systems on Agriculture and Health
Editors Janusz Kacprzyk Polish Academy of Sciences Systems Research Institute Warsaw, Poland
Mostafa Ezziyyani Abdelmalek Essaâdi University Tangier, Morocco
Valentina Emilia Balas Department of Automatics and Applied Software Aurel Vlaicu University of Arad Arad, Romania
ISSN 2367-3370 ISSN 2367-3389 (electronic) Lecture Notes in Networks and Systems ISBN 978-3-031-35247-8 ISBN 978-3-031-35248-5 (eBook) https://doi.org/10.1007/978-3-031-35248-5 © The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publisher, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Switzerland AG The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Foreword
Within the framework of the International Initiative for Sustainable Development of innovations and scientific research in order to keep pace with the digital transformation in light of the fourth industrial revolution and to encourage development projects known to the world, ENSAM-Rabat of Mohammed V University in cooperation with ICESCO organized the fourth edition of the International Conference on Advanced Smart Systems for Sustainable Development and their applications in various fields through five specialized seminars during the period from May 22 to 28, 2022. The fourth edition of the International Conference on Advanced Smart Systems for Sustainable Development was a great success, under the high patronage of His Majesty King of Morocco, Mohammed VI, and the participation of scientists and experts from more than 36 countries around the world. The conference, in its fourth edition, also resulted in a set of agreements and partnerships that were signed between the various participating parties, thus contributing to achieving the goals set by the conference regarding the investment of smart systems for sustainable development in the sectors of education, health, environment, agriculture, industry, energy, economy and security. In view of the importance of the conference as a high-level annual forum, and in consideration of the scientific status that the conference enjoys nationally, continually and internationally. Based on the experience gained and accumulated through the previous editions, we look forward to the success of next edition at all organizational and scientific levels, like its predecessors, and hosting a distinguished presence and weighty personalities from all participating countries in order to move forward for cooperation in priority areas and common interest such as health, agriculture, energy and industry.
Preface
Science, technology and innovation have for a long time been recognized as one of the main drivers behind productivity increases and a key long-term lever for economic growth and prosperity. In the context of the International Conference on Advanced Intelligent Systems for Sustainable Development plays an even more central role. Actually, AI2SD features strongly in Sustainable Development Goal in different fields, as well as being a cross-cutting one to achieve several sectoral goals and targets: Agriculture, Energy, Health, Environment, Industry, Education, Economy and Security. An ambition of the AI2SD to become the global forerunner of sustainable development should, in particular, include integrating new technologies and artificial intelligence and smart systems in its overarching and sectoral strategies of research and development. In which it emphasizes that solutions discussed by experts are important drivers for researches and development. AI2SD is an interdisciplinary international conference that invites academics, independent scholars and researchers from around the world to meet and exchange the latest ideas and discuss technological issues concerning all fields Social Sciences and Humanities for Sustainable Development. Due to the nature of the conference with its focus on innovative ideas and developments, AI2SD provides the ideal opportunity to bring together professors, researchers and high education students of different disciplines, to discuss new issues, and discover the most recent developments, scientific researches proposing the panel discussion on Advanced Technologies and Intelligent Systems for Sustainable Development Applied to Education, Agriculture, Energy, Health, Environment, Industry, Economy and Security.
Organization
Chairs General Chairs Mostafa Ezziyyani Janusz Kacprzyk Valentina Emilia Balas
Abdelmalek Essaadi University, FST – Tangier, Morocco Polish Academy of Sciences, Poland Aurel Vlaicu University of Arad, Romania
Co-chairs Khalid El Bikri Wajih Rhalem Loubna Cherrat Omar Halli
ENSAM Rabat, Morocco ENSAM Rabat, Morocco ENCG of Tangier, Morocco Advisor to the Director General of ICESCO
Honorary Presidents Salim M. Almalik
Abdellatif Miraoui Younes Sekkouri Ghita Mezzour
Director General (DG) of the Islamic World Educational, Scientific and Cultural Organization (ICESCO) Minister of Higher Education, Scientific Research and Professional Training of Morocco Minister of Economic Inclusion, Small Business, Employment and Skills Minister Delegate to the Head of Government in Charge of Digital Transition and Administration Reform
Honorary Guests Thomas Druyen
Jochen Werner
Director and Founder of the Institute for Future Psychology and Future Management, Sigmund Freud University Medical Director and CEO, Medicine University of Essen, Germany
x
Organization
Ibrahim Adam Ahmed El-Dukheri Director General of the Arab Organization for Agricultural Development Stéphane Monney Mouandjo Director General of CAFRAD Jamila El Alami Director of the CNRST Rabat, Morocco Mostapha Bousmina President of the EuroMed University of Fez, Fez, Morocco Chakib Nejjari President of the Mohammed VI University of Health Sciences Casablanca, Morocco Noureddine Mouaddib President of International University of Rabat, Rabat, Morocco Azzedine Elmidaoui President of Ibn Tofail University, Kenitra, Morocco Lahcen Belyamani President of the Moroccan Society of Emergency Medicine SAMU Rabat, Morocco Karim Amor President of Moroccan Entrepreneurs and High Potentials of the World-CGEM Hicham El Abbadi Business Sales Manager, Afrique Francophone EPSON Ilham Berrada Director of ENSIAS Rabat, Morocco Mostafa Stito Director of the ENSA of Abdelmalek Essaadi University, Tetouan, Morocco Mohamed Addou Dean of FST Tangier, Morocco Ahmed Maghni Director of ENCG Tangier, Morocco
Keynote Speakers Chakib Nejjari Anas Doukkali Thomas Druyen
Jochen Werner Abdelhamid Errachid El Salhi Oussama Barakat Fatima Zahra Alaoui Issame Outaleb Rachid Yazami
President of the Mohammed VI University of Health Sciences Casablanca, Morocco Former Minister of Health, Morocco Director and Founder of the Institute for Future Psychology and Future Management Sigmund Freud University Medical Director and CEO, Medicine University of Essen, Germany Full Professor Class Exceptional Class, University Claude Bernard, Lyon, France University of Franche-Comté, Besançon, France Dean of the Faculty of Medicine of Laâyoune, Morocco CEO and Founder PharmaTrace, Munich, Germany Scientist, Engineer and Inventor, Morocco
Organization
Tarkan Gürbüz Plamen Kiradjiev Abdel Labbi
Mostafa Ezziyyani Ghizlane Bouskri Levent Trabzon Marius M. Balas Afef Bohli
Ahmed Allam (President)
Valentina Emilia Balas Faissal Sehbaoui Jaime Lloret Hanan Melkaoui Issa Mouhamed Hossana Twinomurinzi Abdelhafid Debbarh Hatim Rhalem Faeiz Gargouri (Vice President) Adil Boushib Nasser Kettani
Kaoutar El Menzhi Khairiah Mohd-Yusof (President) Nadja Bauer Badr Ikken Amin Bennouna Mohamed Essaaidi Hamid Ouadia
xi
Middle East Technical University (METU), Ankara, Turkey German Edge Cloud (GEC), Friedhelm Loh Group, Germany Head of Data & AI Platforms Research, IBM Distinguished Engineer, IBM Research – Europe FST – Tangier, Morocco Senior Data Scientist at Volkswagen Group, Germany Mechanical Engineering, Istanbul Technical University, Turkey Aurel Vlaicu University of Arad Assistant Professor at the Higher Institute of Computer Science and the Cofounder of Digi Smart Solutions World Association for Sustainable Development, Senior Policy Fellow, Queen Mary University of London, UK Aurel Vlaicu University of Arad, Romania CEO of AgriEDGE, Attached to the Mohammed VI Polytechnic University Department of Communications Polytechnic University of Valencia, Spain Yarmouk University, Irbid, Jordan Head|Centre for Applied Data Science at University of Johannesburg, South Africa Chief of Staff/Advisor to the President-UIR EPSON Sales Manager, Morocco University of Sfax, Tunisia Regional Manager Microsoft, Germany Entrepreneur, ExO Coach, Digital Transformation Expert, Exponential Thinker, Certified DPO, Accessibility Expert Head of Digital Learning Center UM5R, Morocco Johor Bahru, Johor, Malaysia Dortmund, Germany General Director of IRESEN, Rabat, Morocco Cadi Ayyad University, Marrakech, Morocco ENSIAS, Mohammed V University, Rabat, Morocco ENSAM, Mohammed V University, Rabat, Morocco
xii
Organization
Khalid Zinedine Brahim Benaji Youssef Taher Tarik Chafik Abdoulkader Ibrahim Idriss Loubna Cherrat Laila Ben Allal Najib Al Idrissi
Hassan Ghazal Muhammad Sharif Mounir Lougmani El Hassan Abdelwahid Mohamed Zeriab Es-Sadek Mustapha Mahdaoui M’Hamed Ait Kbir Mohammed Ahachad
Faculty of Sciences, Mohammed V University, Rabat, Morocco ENSAM, Mohammed V University, Rabat, Morocco Center of Guidance and Planning of Education, Morocco FST, Abdelmalek Essaadi University, Tangier, Morocco Dean of Faculty of Engineering – University of Djibouti, Djibouti Abdelmalek Essaadi University, Morocco FST Abdelmalek Essaadi University, Morocco Mohammed VI University of Health Sciences, General Secretary of the Moroccan Society of Digital Health, Morocco President of the Moroccan Association of Telemedicine and E-Health, Morocco Director and Founder of Advisor/Science and Technology at ICESCO General Secretary of the Association of German Moroccan Friends-DMF Cadi Ayyad University, Marrakech ENSAM, Mohammed V University in Rabat FST, Abdelmalek Essaadi University, Morocco Abdelmalek Essaadi University, Morocco Abdelmalek Essaadi University, Morocco
Course Leaders Adil Boushib Ghizlane Bouskri Nadja Bauer Hassan Moussif Abdelmounaim Fares Imad Hamoumi Ghizlane Sbai
Regional Manager Microsoft, Germany Senior Data Scientist at Volkswagen Group, Germany Dortmund, Germany Deutsche Telekom expert, Germany. General Director and Founder of M-tech Co-Founder and Chief Executive Officer Guard Technology, Germany Senior Data Scientist Engineer, Germany Product Owner, Technical Solution Owner at Pro7Sat1
Organization
Scientific Committee Christian Axiak, Malta Bougdira Abdeslam, Morocco Samar Kassim, Egypt Vasso Koufi, Greece Alberto Lazzero, France Charafeddine Ait Zaouiat, Morocco Mohammed Merzouki, Morocco Pedro Mauri, Spain Sandra Sendra, Spain Lorena Parra, Spain Oscar Romero, Spain Kayhan Ghafoor, China Jaime Lloret Mauri, Spain Yue Gao, UK Faiez Gargouri, Tunis Mohamed Turki, Tunis Abdelkader Adla, Algeria Souad Taleb Zouggar, Algeria El-Hami Khalil, Morocco Bakhta Nachet, Algeria Danda B. Rawat, USA Tayeb Lemlouma, France Mohcine Bennani Mechita, Morocco Tayeb Sadiki, Morocco Mhamed El Merzguioui, Morocco Abdelwahed Al Hassan, Morocco Mohamed Azzouazi, Morocco Mohammed Boulmalf, Morocco Abdellah Azmani, Morocco Kamal Labbassi, Morocco Jamal El Kafi, Morocco Dahmouni Abdellatif, Morocco Meriyem Chergui, Morocco El Hassan Abdelwahed, Morocco Mohamed Chabbi, Morocco Mohamed_Riduan Abid, Morocco Jbilou Mohammed, Morocco Salima Bourougaa-Tria, Algeria Zakaria Bendaoud, Algeria Noureddine En-Nahnahi, Morocco Mohammed Bahaj, Morocco Feddoul Khoukhi, Morocco Ahlem Hamdache, Morocco
xiii
xiv
Organization
Mohammed Reda Britel, Morocco Houda El Ayadi, Morocco Youness Tabii, Morocco Mohamed El Brak, Morocco Abbou Ahmed, Morocco Elbacha Abdelhadi, Morocco Regragui Anissa, Morocco Samir Ahid, Morocco Anissa Regragui, Morocco Frederic Lievens, Belgium Emile Chimusa, South Africa Abdelbadeeh Salem, Egypt Mamadou Wele, Mali Cheikh Loukobar, Senegal Najeeb Al Shorbaji, Jordan Sergio Bella, Italy Siri Benayad, Morocco Mourad Tahajanan, Morocco Es-Sadek M. Zeriab, Morocco Wajih Rhalem, Morocco Nassim Kharmoum, Morocco Azrar Lahcen, Morocco Loubna Cherrat, Morocco Soumia El Hani, Morocco Essadki Ahmed, Morocco Hachem El Yousfi Alaoui, Morocco Jbari Atman, Morocco Ouadi Hamid, Morocco Tmiri Amal, Morocco Malika Zazi, Morocco Mohammed El Mahi, Morocco Jamal El Mhamdi, Morocco El Qadi Abderrahim, Morocco Bah Abdellah, Morocco Jalid Abdelilah, Morocco Feddi Mustapha, Morocco Lotfi Mostafa, Morocco Larbi Bellarbi, Morocco Mohamed Bennani, Morocco Ahlem Hamdache, Morocco Mohammed Haqiq, Morocco Abdeljabbar Cherkaoui, Morocco Rafik Bouaziz, Tunis Hanae El Kalkha, Morocco Hamid Harroud, Morocco
Organization
Joel Rodrigues, Portugal Ridda Laaouar, Algeria Mustapha El Jarroudi, Morocco Abdelouahid Lyhyaoui, Morocco Nasser Tamou, Morocco Bauer Nadja, Germany Peter Tonellato, USA Keith Crandall, USA Stacy Pirro, USA Tatiana Tatusova, USA Yooseph Shibu, USA Yunkap Kwankam, Switzerland Frank Lievens, Belgium Kazar Okba, Algeria Omar Akourri, Morocco Pascal Lorenz, France Puerto Molina, Spain Herminia Maria, Spain Driss Sarsri, Morocco Muhannad Quwaider, India Mohamed El Harzli, Morocco Wafae Baida, Morocco Mohammed Ezziyyani, Morocco Xindong Wu, China Sanae Khali Issa, Morocco Monir Azmani, Morocco El Metoui Mustapha, Morocco Mustapha Zbakh, Morocco Hajar Mousannif, Morocco Mohammad Essaaidi, Morocco Amal Maurady, Morocco Ben Allal Laila, Morocco Ouardouz Mustapha, Morocco Mustapha El Metoui, Morocco Said Ouatik El Alaoui, Morocco Lamiche Chaabane, Algeria Hakim El Boustani, Morocco Azeddine Wahbi, Morocco Nfaoui El Habib, Morocco Aouni Abdessamad, Morocco Ammari Mohammed, Morocco El Afia Abdelatif, Morocco Noureddine En-Nahnahi, Morocco Zakaria Bendaoud, Algeria Boukour Mustapha, Morocco
xv
xvi
Organization
El Maimouni Anas, Morocco Ziani Ahmed, Morocco Karim El Aarim, Morocco Imane Allali, Morocco Mounia Abik, Morocco Barrijal Said, Morocco Mohammed V., Rabat, Morocco Franccesco Sicurello, Italy Bouchra Chaouni, Morocco Charoute Hicham, Morocco Zakaria Bendaoud, Algeria Ahachad Mohammed, Morocco Abdessadek Aaroud, Morocco Mohammed Said Riffi, Morocco Abderrahim Abenihssane, Morocco Abdelmajid El Moutaouakkil, Morocco Silkan, Morocco Khalid El Asnaoui, France Salwa Belaqziz, Morocco Khalid Zine-Dine, Morocco Ahlame Begdouri, Morocco Mohamed Ouzzif, Morocco Essaid Elbachari, Morocco Mahmoud Nassar, Morocco Khalid Amechnoue, Morocco Hassan Samadi, Morocco Mohammed Yahyaoui, Morocco Hassan Badir, Morocco Ezzine Abdelhak, Morocco Mohammed Ghailan, Morocco Kaoutar Elhari, Morocco Mohammed El M’rabet, Morocco El Khatir Haimoudi, Morocco Mounia Ajdour, Morocco Lazaar Saiida, Morocco Mehdaoui Mustapha, Morocco Zoubir El Felsoufi, Morocco Khalil El Hami, Morocco Yousef Farhaoui, Morocco Mohammed Ahmed Moammed Ail, Sudan Abdelaaziz El Hibaoui, Morocco Othma Chakkor, Morocco Abdelali Astito, Morocco Mohamed Amine Boudia, Algeria Mebarka Yahlali, Algeria
Organization
Hasna Bouazza, Algeria Zakaria Bendaoud, Algeria Naila Fares, Spain Brahim Aksasse, Morocco Mustapha Maatouk, Morocco Abdel Ghani Laamyem, Morocco Abdessamad Bernoussi, Morocco
xvii
Acknowledgement
This book is the result of many efforts combined with subtle and strong contributions more particularly from the General Chair of AI2SD’2022 Professor Mostafa EZZIYYANI from Adelmalek Essaadi University, the distinguished honorary Chair Academician Janusz KACPRZYK from the Polish Academy of Sciences, and Co-Chair Professor Valentina EMILIA BALAS, Aurel Vlaicu University of Arad, ROMANIA. The scientific contribution published throughout this book could never be so revolutionary without the perpetual help and the limitless collaboration of several actors who supreme is precisely the high patronage of his majesty King Mohammed VI, who in addition to his undeniable support in all the production and scientific inspiration processes, he provided us with all the logistical and technical means in the smallest needs presented during the organization of the event and the publication of this book. The deep acknowledgment addressed to ENSAM school embodied by its director Pr. Khalid BIKRI for his prestigious inputs and the valuable contributions provided by Pr. Wajih RHALEM and by all the faculty members and his engineering students have prepared a fertile ground for presentation and exchange resulting in rigorous articles which are published in this volume. Great thanks to the Director General of the Organization of the Islamic World for Education, Science, and Culture (ICESCO) presented by its Director General Dr. Salim M. Al MALIK for their collaboration, their support, and for the distinguished welcome of the researchers and guests from the AI2SD’2022 conference. The appreciation is addressed to Dr. Omar HALLI advisor of the Director General of ICESCO for His excellent role in coordinating the organization of the AI2SD’2022 edition at ICESCO. The dedication inevitably concerns the organizing committee managed by General Chair Professor Mostafa EZZIYYANI, the VIP coordinator Professor Mohammed Rida ECH-CHARRAT, the scientific committee coordinator Professor Loubna CHERRAT, the Ph.D. student organization committee coordinator Mr. Abderrahim EL YOUSSEFI, and all professors and doctoral students for their constant efforts for the organization, maintenance of the relationship with researchers and collaborators, and also in the publication process
Contents
Breast Cancer Progression Prediction for Care Treatment Efficiency and Intelligent Adaptation, Based on AI Algorithm Classification . . . . . . . . . . . . Sarah Khrouch, Maroi Tsouli Fathi, Abderrahim El Yessefi, Loubna Cherrat, Wajih Rhalem, and Mostafa Ezziyyani Study of the Germination of Wild and Cultivated Blackberries of the Northern Region of Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Amal Maurady, Malika M’guil, Dounia Harama, Iman Touati, Naima Bel Mokhtar, Soumaya El Ismaili, Leila Karimi, Mohammed Reda Britel, and Ahlam Hamim
1
8
The Impact of Covid 19 on Recommendation Platforms . . . . . . . . . . . . . . . . . . . . . Amina Samih, Abderrahim Ghadi, and Abdelhadi Fennan
19
Classifiers-Based Personality Disorders Detection . . . . . . . . . . . . . . . . . . . . . . . . . . Fatemeh Sajadi Ansari, Djamal Benslimane, Aymen Khelifi, and Mahmoud Barhamgi
35
How Health Information Technology Improved Patient Care and Treatment During the COVID-19 Pandemic: A Comparison Between International Case Studies and the Moroccan Context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Ahmed Kadiri, Hamid Azzouzi, and Noufel Sefiani
46
Dates Detection System for Automatic Harvesting Using Deep Learning . . . . . . Yousra Zarouit, Brahim Aksasse, and Mohamed Ouhda
58
Machine Learning for Diabetes Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Sara Retal, Hajar Sahbani, Nassim Kharmoum, Wajih Rhalem, and Mostafa Ezziyyani
75
Technology as an Answer to the Trust Crisis in Mental Health Services Digitization Serving Mental Health Care Systems . . . . . . . . . . . . . . . . . . . . . . . . . . El Mehdi Ghalim and Abdelmajid Elouadi Artificial Intelligence at the Service of Precision Medicine . . . . . . . . . . . . . . . . . . Wafae Abbaoui, Sara Retal, Nassim Kharmoum, and Soumia Ziti
84
91
Case-Based Reasoning Approach, Integrating Deep Learning for Patient Diagnosis Combined X-Ray with Symptoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Moulay Youssef Ichahane, Noureddine Assad, and Hassan ouahmane
xxii
Contents
AGRI-PREDI Prediction System of Climate Change Based on Machine Learning for Precision Agriculture in Mediterranean Region . . . . . . . . . . . . . . . . . 116 Maroi Tsouli Fathi, Ramz Tsouli Fathi, Sarah Khrouch, Loubna Cherrat, and Mostafa Ezziyyani Estimated Daily Reference Evapotranspiration Using Machine Learning and Deep Learning Based on Various Combinations of Meteorological Data . . . 128 Ayoub Ba-ichou, Abderrahim Waga, Ali Bekri, and Said Benhlima Deep Learning Approach for Brain Tumor Classification Implemented in Raspberry Pi . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 Nihal Remzan, Karim Tahiry, and Abdelmajid Farchi Machine Learning for Medical Image Analysis: A Survey . . . . . . . . . . . . . . . . . . . 148 Amina Fettah, Rafik Menassel, and Abdeljalil Gattal Simulation Study on Validating an Efficient Technique for Early Breast Cancer Detection Using Thermographic Micro-biosensors . . . . . . . . . . . . . . . . . . . 165 Zakaryae Khomsi, Achraf Elouerghi, and Larbi Bellarbi Argania Forest Change Detection from Sentinel-2 Satellite Images Using U-Net Architectures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174 Soufiane Idbraim, Taha Bouhsine, Mohamed Reda Dahbi, Antoine Masse, and Manuel Arbelo Development of Environmental Performance Measurement Model for Public Hospitals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185 Hajar Regragui, Naoufal Sefiani, Hamid Azzouzi, and Kamal Reklaoui The Nitrogen Effects on Growth and Development of Morphological Parameters of “Argania Spinosa L. skeel” Tree Seedlings . . . . . . . . . . . . . . . . . . . 198 Hassania Farhoune and Souad Cherkaoui Cost-Effective Manufacturing Operations During and After the COVID-19 Pandemic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 Karim Haricha, Azeddine Khiat, Yassine Issaoui, Ayoub Bahnasse, and Hassan Ouajji Hypertension Risk Prevention: Towards a Predictive Model Based on Data Mining . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234 Manar Aarrad, El Mostafa Rajaallah, and Mohamed Hilal A Lightweight CNN Model for Tomato Crop Disease Detection . . . . . . . . . . . . . . 242 Mohamed Lmoussaoui, El Hassan Ait Laasri, Abderrahman Atmani, and Driss Agliz
Contents
xxiii
Discrete Event Simulation for the Evaluation of Emergency Department Layout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 Khalil Bouramtane, Said Kharraja, Jamal Riffi, and Omar El Beqqali Word Sense Disambiguation in the Biomedical Domain: Short Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258 Oumayma El Hannaoui, El Habib Nfaoui, and Fatima El Haoussi Environmental Impact Assessment of Agricultural Practices Using the Life Cycle Assessment Method: A Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272 Imane Mehdi, Aicha Bouchara, Mohammed Ammari, and Laïla Ben Allal Weakly Supervised Patchwise CT Liver Segmentation . . . . . . . . . . . . . . . . . . . . . . 284 Youssef Ouassit, Soufiane Ardchir, Mohamed Yassine El Ghoumari, Mohamed Bouni, and Mohamed Azzouazi Smart Agriculture Using an Enhanced Selection for MPRs . . . . . . . . . . . . . . . . . . 294 Ayoub Abdellaoui Prevalence of Anxiety and Depressive Disorders and Cognitive Affective Risk Factors: Case of Officials of the Rabat-Sale-Kenitra Region . . . . . . . . . . . . . 305 E. Drissi, S. Boulbaroud, H. Hami, A. Ahami, and F.-Z. Azzaoui Predicting Drug Compounds Effectiveness Based on Chemical Properties and Bioactivity Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318 Hamza Hanafi, Badr Dine Rossi Hassani, and M’hamed Aït Kbir A Review of the Transportation Routing Problem During the COVID-19 Pandemic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327 Anouar Annouch and Adil Bellabdaoui The Effect of a Physical Activity Program on Improving Sports Performance in Obese Adolescents Based on a Statistical Analysis of Cardiovascular Physiological Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343 Youssef El Machrouh, El Mostafa Rajaallah, and Mustapha Krim HER-Omics, a Model of Transcriptomics Data Integration in EHRs . . . . . . . . . . . 349 Nihal Habib, Abdellah Idrissi Azami, Douae El Ghoubali, Zainab El Ouafi, Mustapha Lemsayah, Abdesselam Bougdira, Najib Al Idrissi, Wajih Rhalem, Mostafa Ezziyyani, Chakib Nejjari, and Hassan Ghazal
xxiv
Contents
Artificial Intelligence for Stroke Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359 Jamal Gsim, Mohamed Zeriab Es-sadek, Wajih Rhalem, Nassim Kharmoum, Najib Al idrissi, Lahcen Belyamani, Amal Rami, Jehane Aasfara, Saïd Jidane, Mostafa Ezziyyani, and Hassan Ghazal The Role of Uncertainty Avoidance, Trust, and Land Tenure in Predicting the Adoption of Green IoT Irrigation Systems in Morocco: An Improved Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368 Zitan Houda and Chafik Khalid The Complete Mitochondrial Genome of Agapornis roseicollis . . . . . . . . . . . . . . 384 Douae El Ghoubali, Stacy Pirro, Abdellah Idrissi Azami, Sofia Sehli, Chaimaa Berrahma, Najib Al Idrissi, Wajih Rhalem, Chakib Nejjari, and Hassan Ghazal Cross-Talk Between Intramolecular and Intermolecular Amino Acid Networks Orchestrates the Assembly of the Cholera Toxin B Pentamer via the Residue His94 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391 Mounia Achoch, Giovanni Feverati, Kave Salamatian, Laurent vuillon, and Claire Lesieur Identification of Responsibilities for the Use of Devices Incorporating Artificial Intelligence in Health: Systemic Review . . . . . . . . . . . . . . . . . . . . . . . . . . 405 M. Qouhafa, B. Benaji, S. Lebbar, A. Soulaymani, A. Moukhtari, M.H Elyoussfialaoui, and B. Nsiri Artificial Intelligence Recognition of Human Body Actions for Bionic Applications (New Combination of Indicator Parameters) . . . . . . . . . . . . . . . . . . . 416 Oussama Lamsellak, Ahmad Benlghazi, Abdelaziz Chetouani, Abdelhamid Benali, and Driss Mousaid Parkinson’s Disease Recognition from Speech Signal Using Discrete Wavelet Transform, Delta, Delta-Delta, and K-Nearest Neighbor . . . . . . . . . . . . . 426 N. Boualoulou, T. Belhoussine Drissi, and B. Nsiri Triple Negative Breast Cancer and Non-Triple Negative Breast Cancer Recurrence Prediction Using Boosting Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440 Saadia Azeroual, Fatima-ezzahraa Ben-Bouazza, Amine Naqi, and Rajaa Sebihi A Word-Based Moroccan Sign Language Dataset for Emergency Situations: Use Case COVID-19 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451 Ilham El Ouariachi, Rachid Benouini, Khalid Zenkouar, and Arsalane Zarghili
Contents
xxv
The Use of ISFET for the Measurement of Phosphorus in Moroccan Soils . . . . . 462 Oumayma Benslimane, Reda Rabie, and Souad El Hajjaji Exploring the Performance of Traditional Word Embeddings and Contextual Embedding Models for COVID-19 Sentiment Analysis . . . . . . . . 469 Nouhaila Bensalah, Habib Ayad, Abdellah Adib, and Abdelhamid Ibn El Farouk Mycoflora of Dormant Crocus Sativus Corms in Morocco . . . . . . . . . . . . . . . . . . . 479 Samah Ourras, Ismail E L Aymani, Najoua Mouden, Karima Selmaoui, Soukaina Msairi, M’hammed Elouark, Rachid Benkirane, Cherkaoui E L Modafar, Amina Ouazzani Touhami, and Allal Douira Optimized DL-Based Model for Hypertrophic Cardiomyopathy CMR Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490 Sara El Omary, Souad Lahrache, and Rajae El Ouazzani Low-Cost Water Conductivity Sensor Based on a Parallel Plate Capacitor for Precision Agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500 Sandra Sendra, Sandra Viciano-Tudela, Alberto Ivars-Palomares, and Jaime Lloret Literature Review of Deep Learning for Tuberculosis Based on Chest Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515 Lahlou Sara and Ziti Soumia Inventory of Corticolous Lichens and Argan Wood Fungi . . . . . . . . . . . . . . . . . . . 521 Abdesslam Bouchar, Zineb Sellal, Soukaina Maazouzi, Soukaina Msairi, Saifeddine El Kholfy, Rachid Benkirane, Amina Ouazzani Touhami, and Allal Douira Detection of Alpha-Cypermethrin in Food Using a Sensor Array Coupled to Machine Learning Algorithms: The Case of the Wild Edible Swiss Chard . . . 528 Ali Amkor and Noureddine El Barbri Effect of Some Derivatives of Pyridazin-3 (2h) – Ones on the in Vitro and in Situ Development of Different Pathogenic Fungi on Citrus Fruits . . . . . . . 536 D. Boudoudou, S. El Marrakchi, A. Talha, B. Kerroum, A. Ouazzani Touhami, A. Douira, and H. Benyahia Recovery of Organic Waste by Biogas Production-Mathematical Modeling of Anaerobic Digestion: A Short Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . 552 Youssef Benyahya, Mohamed Sadik, and Abderrahim Fail
xxvi
Contents
Automation of Agriculture Using Artificial Intelligence: Towards a Sustainable Agriculture in Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566 Rachid Batess, Younes El Fellah, Reda Errais, Ghizlane Bouskri, and El Houssain Baali Review of Weed Detection Methods Based on Machine Learning Models . . . . . . 576 Bouchra El Jgham, Otman Abdoun, and Haimoudi El Khatir Analysis of Alcoholic EEG Signals Based on Discrete to Continuous Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587 Hayat Sedrati, Wajih Rhalem, Nabil Aqili, Mohamed Zeriab Es-Sadek, Mostafa Ezziyani, Sory Millimono, Nassim Kharmoum, Fatima El Omari, Chakib Nejjari, and Hassan Ghazal Growth Promoting of Tomato Plants by Incorporation of Trichoderma asperellum Enriched Liquid Product via Foliar Spray and the Irrigation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599 Mouden Najoua, Ouazzani Touhami Amina, Albatnan Abdelmoti, Selmaoui Karima, Benkirane Rachid, and Douira Allal Effect of Trichoderma Asperellum on the Development of Strawberry Plants and Biocontrol of Anthracnose Disease Caused by Colletotrichum Gloeosporioides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 609 Hanane El Kaissoumi, Fadoua Berber, Najoua Mouden, Abdelatif Ouazzani Chahdi, Abdelmoti Albatnan, Amina Ouazzani Touhami, Karima Selmaoui, Rachid Benkirane, and Allal Douira Diversity of Arbuscular Mycorrhizal Fungi in the Rhizosphere of Four Citrus Genotypes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623 Manal Bourazza, Ouiam Chetto, Abdelhak Talha, Allal Douira, and Hamid Benyahia Contribution to the Development of a Technological Platform for Analysis in Precision Agriculture for the Biovigilance of Cryptogamic Diseases in Strawberry (Fragaria × ananassa) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 637 Mohammed Ezziyyani, Ahlem Hamdache, Loubna Cherrat, Ikram Laksiby, Mostafa Ezziyyani, Hakim Alilou, Jaime Mauri, Asma Chaik, and Catalina Egea Gilabert Diversity of the Bryoflora of the Tazekka National Park, North-East Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 645 Saadi Kamal, Achoual Khalid, El Khaddari Amal, Zidane Lahcen, Dahmani Jamila, and Belahbib Nadia
Contents
xxvii
A New Model-Based Approach for Migrating Health 2.0 to Health 3.0 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673 Mint Mohamed Meyma, Naziha Laaz, and Samir Mbarki Effect of the Impact of Environmental Factors on the Diet of Horse Mackerel (Trachurus trachurus) from the Coasts of the Moroccan . . . . . . . . . . . . 683 Badri Maroua, Benchoucha Said, Achagra Chaima, Asma Chaik, and Ezziyyani Mohammed Study of the Risks Associated with the Use of Pesticides on Soils and Groundwater in the Loukkos Perimeter (Morocco) . . . . . . . . . . . . . . . . . . . . . . 693 Bagayou Ahmed, Hamdache Ahlem, Diane Yassine, and Ezziyyani Mohammed Diversity of the Bryoflora of a Cedar Forest in the Central High Rif, Northern Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 700 Khalid Achoual, Aomar Dabghi, Kamal Saadi, Houria El Ouahdani, Ahmed Bouhssini, Jamila Dahmani, and Nadia Belahbib Toward a Smart Agriculture Using Deep Learning for Plant Disease Detection: Challenges and Opportunities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725 Abdelouafi Boukhris, Hiba Asri, and Antari Jilali Study of the Effect of Plastic Film in Tunnels on the Yield and Quality of Strawberry Plants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 735 Mesbahi Imane, Hamim Ahlam, Hamdache Ahlem, El Moudden Manal, and Ezziyyani Mohammed Effect of Endomycorrhizae on the Pathogenicity of Rhizoctonia solani and the Growth of Olive Plants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743 Soukaina Msairi, Mohamed Chliyeh, Soumaya El Gabardi, Abdelaziz El Alaoui Moulay, Karima Selmaoui, Afifa Mouria, Rachid Benkirane, Amina Ouazzani Touhami , and Allal Douira Detection of Alzheimer’s Disease Using a Convolutional Neural Network . . . . . 757 Rachid Alhyane, Aziza Kassimi El Bakkali, Abdelaziz Bouroumi, Florence Rémy, and Abdelhakim El Boustani Distribution of Resellers and Modeling of the Future of the Majority Pesticides for Agricultural Use in Morocco . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 766 Mohamed Mchita, Ahlem Hamdache, Asma Chaik, and Mohammed Ezziyyani
xxviii
Contents
A Proposed Architecture Based on Deep Learning and Optimization Techniques for Medical Diagnostic Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 773 Ibtihal Mouhib, Meryeme Hadni, Manal El Bajta, and Hassan Naanani A Hierarchical Machine Learning Algorithm for Epileptic Seizure Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 785 Mohamed Abdelbaki, Charafeddine Aitzaouiat, Habiba Elfatouaki, and Adnane Latif Epileptic Seizure Prediction Using Artificial Intelligence Methods . . . . . . . . . . . . 792 Ilyas Zidane, Jamal Mhamdi, Mostafa Ezziyyani, Wajih Rhalem, and Nordine Zidane New Approach of 3D Protein Structure Superimposition: Case Study of “SARS-COV-2” and “SARS-COV” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 805 Nabil Aqili, Wajih Rhalem, Mohamed Zeriab Es-sadek, Hayat Sedrati, Najib alidrissi, Saïd Jidane, Imane Allali, Salsabil Hamdi, Zainab Elouafi, Nassim Kharmoum, Mostafa Ezziyani, Lahcen Belyamani, and Hassan Ghazal Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application in Biomedicine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 816 Faoussi Mohamed, Tbib Bouazza, and Bounou Salim Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization and Advanced Structure Study Application: Improve Mechanical Strength of Implant and the Osteointegration Process . . . . . . . . . . . . 829 Faoussi Mohamed, Tbib Bouazza, Zakaria Kbiri, and Bounou Salim Extraction and Characterization of New Cellulosic Fibers from Moroccan Mallow Stem and Comparison with Other Naturel Fibers . . . . . . . . . . . . . . . . . . . . 845 Youssef El Omari, Bouazza Tbib, Mohammed Eddya, Zakaria Kbiri, and Khalil El-Hami Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 855
Breast Cancer Progression Prediction for Care Treatment Efficiency and Intelligent Adaptation, Based on AI Algorithm Classification Sarah Khrouch1(B) , Maroi Tsouli Fathi1 , Abderrahim El Yessefi1 , Loubna Cherrat2 , Wajih Rhalem3 , and Mostafa Ezziyyani1 1 Mathematics and Applications Laboratory, Faculty of Sciences and Techniques of Tangier,
Abdelmalek Essaâdi University, Tangier, Morocco [email protected], [email protected], [email protected], [email protected] 2 National School of Commerce and Management Tangier, Abdelmalek Essaâdi University, Tangier, Morocco [email protected] 3 National School of Arts and Crafts of Rabat, Mohammed V University of Rabat, Rabat, Morocco [email protected]
Abstract. In recent years death rates is extremely increasing due to breast cancer. Currently, it is the most commonly diagnosed malignancy among women around the globe. Despite the studies made early detection of breast cancer there is a need for predicting factors and biomarkers of treatment response. Early monitoring of the effectiveness of treatment can thus recognize non responding tumors and provide to adapt the therapy and select the most beneficial therapy for individual patient, also it helps to calculate the improvable toxicity. Our main goal is to make a significant contribution toward improving the quality of healthcare. This work strives to create a predictive model to predict the eventual therapeutic response of breast cancer patients after first cycle of treatment which could help the doctors to select easier the suitable therapeutic protocol using machine learning methods. Machine learning techniques are more used and are extremely promising in medicine field and can improve the treatment process and predict treatment outcomes. This paper describes briefly the recent applications and various machine learning algorithms including SVM, random forest, logistic regression, and NN, and their use in our study to predict treatment response based on synthetic data. Keywords: Breast cancer · Machine learning · Treatment response · Therapeutic protocol · Biomarkers · Predictive model
1 Introduction Breast cancer is the first and most common disease of the human genome in women, highly complex and driven by a multitude of genetic and epigenetic factors, no one can © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 1–7, 2023. https://doi.org/10.1007/978-3-031-35248-5_1
2
S. Khrouch et al.
deny that cancer is one of the major health problems worldwide. The statistics have shown that 8.8 million people worldwide died from cancer in 2015, that is nearly 1 in 6 of all global deaths (WHO) [1]. The World Health Organization WHO has estimated that cancer causes more deaths than all cardiovascular diseases [1]. Cancer is an excessively complex disease and can become a real danger to the survival of the living being. Besides, cancer is a disease caused by a change that arises in genes or by a lesion to genes what’s called a genetic mutation, this disease characterized by the division of abnormal cells that form lumps or growing mass of cancer cells, called a tumor which may invade other normal body tissue. Understanding the mutational profile of a tumor is important to inform diagnosis and guide treatment options, as well as to identify patients who may not respond to treatment. Therefore, early diagnosis and prognosis of this disease are very important to increase the chances of survival of patients. During our research, we found that more than 50,000 new breast cancer cases were reported in 2020, and this disease has caused a significant number of deaths, it is estimated that over 30,000 people will die as a result of breast cancer in Morocco in 2020. The arrival of new information and communication technologies to estimate therapy’s effectiveness according to the profile of the tumor makes it possible to offer patients more personalized treatments and more precisely in cancer care. The use of big data in healthcare has made a big change in BI by allowing for strategic planning thanks to better insights into people’s motivations and analyzing huge amount of healthcare data. Also, big data and data mining play a significant role due to their high performance in predicting, diagnosis of the diseases, Data mining algorithms, not only help analyze medical information but reducing costs of medicine, making real time decision to save people’s lives. Precision medicine uses these technologies to diagnose diseases and tailor treatments to the individual. A biomarker is a characteristic that can be measured as an indicator of normal or pathogenic biological processes and pharmacological responses to a therapeutic intervention. Specific gene modifications and molecular signatures are used as biomarkers. The stakes are high, as the future of the patient in the first metastatic relapse depends in part on the first-line treatment choice. We developed and compared several approaches to develop prediction models for pathological complete response in breast cancer patients This paper mainly gives a comparison between the performance of several classifiers: Support Vector Machine (SVM)), Random Forest, Logistic Regression, and Neural Network (NN). Our objective is to predict the eventual therapeutic response of breast cancer patients after first cycle of treatment, using machine-learning algorithms, and find out the most effective based on the performance of each classifier. The rest of this paper is organized as follows. Section 2 presents Previous research on breast cancer treatment response methods and results. Section 3 the proposed approach. Section 4 the results of the experiments. Section 5 conclusion and future work.
Breast Cancer Progression Prediction
3
2 Related Works Applied machine learning algorithms in the healthcare sector especially in oncology are currently primarily used to analyze large amounts of data to help doctors. Some of the machine learning algorithms commonly used are: Support vector machines (SVM) are a classification method of supervisor type; were developed in the 1990s from the theories of statistical learning, Random Forest Random Forest is a trademarked term for an ensemble of decision trees, is an ensemble learning method for classification, regression and other tasks that operates by constructing a multitude of decision trees at training time. Logistic Regression or logit regression is a statistical model, it is used to estimate real values based on continuous variable(s). Decision tree (C4.5) is a supervised classification algorithm, published by Ross Quinlan. It is based on the ID3 algorithm, is used in Data Mining as a Decision Tree Classifier which can be employed to generate a decision, based on a certain sample of data. NN Neural Network Artificial neural networks are a class of systems inspired by the functioning of biological nervous systems, they imitate their ability to learn from observations and generalize on new situations. Several researchers had realized studies in precision medicine by using many datasets such as using the TCGA dataset, IRM image dataset, METABRIC Dataset, and datasets from various hospitals. These are some significant researches. The author Rachel Choi [2], demonstrates the use of a deep neural network in predicting PCR by using breast MRI imaging and finds out that a deep neural network algorithm is the best and most promising algorithm in personalized prognostic. On the other side, we find that Nicholas Meti [3], worked on a predictive model for an early NAC response using clinical and pathological data. His model was developed and compared with five ML models: k-nearest neighbor classifier, random forest (RF) classifier, naive Bayes algorithm, support vector machine, and multilayer perceptron model, and explains how RF is a classifier that performed best than other machine learning algorithms with an AUC = 0.88, the sensitivity of 70.7%, and specificity of 84.6%. Na Lae Eun [4], found that the random forest model (AUC, 0.82) had better diagnostic performance in showing association with a complete pathological response compared to six other machine learning classifiers (AUC: adaptive boosting, 0.76; decision tree, 0.70; k-nearest neighbor, 0.80; linear support vector machine, 0.75; naive Bayes, 0.74; linear discriminant analysis, 0.79). In recent works, we find that Hyun-Soo Park [5], demonstrated that the random forest, naive Bayes and logistic regression models produced systematically higher AUC values than the decision tree or the ANN model (p < 0.05).
3 Proposed Approach The main objective of our study is to identify the effectiveness of machine learning methods to differentiate between non-responders and strong responders to treatment at an
4
S. Khrouch et al.
early stage of treatment, a predictive algorithm for the treatment response of breast cancer; therefore, we applied machine learning classifiers Support Vector Machine (SVM), Random Forests, Logistic Regression, and NN. Our experiment made based on synthetic data generated. Big data description In our article, we use a synthetic data which contains long-term clinical follow-up data for each patient a clinic biologic data: age at diagnosis[30, 70], Tumor size = [0.02–5] = 2 mm to 5 cm, breast cancer location, histologic type and subtype, Stage [T1, T2, T3, T4], Grad [1–3], Hormone receptors (ER+, ER−), Gene expression, Menopause, metastatic status at diagnosis and location of metastases, treatment received, histologic response to treatment, risk group, disease progression, date of diagnosis, date and status of disease after treatment Cancer. Methodology We begin with generate a synthetic data by selecting the attributes, set target and features extraction. In order to evaluate the performance of data mining algorithm that used in this work. Then we compare their results obtained in order to select the algorithm with the higher accuracy (Figs. 1 and 2). Data set
Normalization
Feature Extraction
Classification
Prediction
Fig. 1. Process of treatment response of breast cancer using machine learning Algorithm
Fig. 2. Approach proposed
Features Extraction 3.1 Assessment of Response to Treatment Assessing tumor response to treatment is based on studying the histologic response. This involves counting the number of viable tumor cells, in order to determine imaging
Breast Cancer Progression Prediction
5
biomarkers to help in medical decision-making and to better understand the biology of breast cancer. RECIST RECIST is the most commonly method used to assess response to treatment. In RECIST, the most important point is the determination of tumor progression in three ways: target lesions, non-target lesions, new lesions [6]. There are some criteria such as tumor size, intensity, shape. 3.2 Radiomic or Computational Medical Is one of recent discipline, its objective, is to better characterize tumors by making indepth use of data directly contained in conventional imaging approaches. In order to extract biological information or features.
4 Result and Discussion Machine learning Algorithms applied to the synthetic data, patients were categorized into two categories R (Responder) and NR (Non-Responder) In order to evaluate and compare the models and identify the best algorithm for the predictive model to predict early treatment response, we use accuracy score. In Table 1, we can see the accuracy percentage for synthetic data. As we can see that SVM has a higher accuracy. Table 1. Performance of the classifier Classifiers
Accuracy (%)
Data
C4.5
95.13%
Synthetic data
SVM
97.13%
RF
95.99%
NN
95.27%
LR
90%
The early prediction of the response in order to help Doctors to avoid ineffective treatment based on synthetic data. The biomarkers or the predictive factors allow for better targeting of drugs using specific and individual criteria for each patient. Therefore, our study sought to identify early predictive factors of response in breast cancer, including clinical pathological parameters (tumor size, grade and type). As we can see in the Fig. 3 shows two categories. The blue color presents responder to treatment after first cycle, the red one presents the patients non responder.
6
S. Khrouch et al.
Fig. 3. Responder/Non-Responder patients’ classification (Color figure online)
5 Conclusion and Perspectives In this papier we see a brief description of breast cancer, provides a study of various technical, we summarizing different algorithms of data mining and we present the research review in the use of data mining. Proposed an architecture of our predictive model and the Machine learning applied for it should be noted that all the results obtained are related just to the synthetic data, it can be considered as a limitation of our work, it is therefore necessary to reflect for future works to apply these same algorithms and methods on real Moroccan case from hospitals to confirm the results obtained via this data and compare them, as well as, in our future works, we plan to apply our and other machine learning algorithms using new parameters on larger data sets with more disease classes to obtain higher accuracy. Acknowledgments. We would like to thank the committee of the International Conference on Advanced Intelligent Systems for Sustainable Development (AI2SD’2022) applied to Health, Agriculture, Energy, Education, and Environment for the opportunity that it offers to participate in this great scientific conference, which opened -for us- the world of scientific research and enabled us to take another step in its path.
References 1. WHO: World health organization 2. Choi, R., Joel, M., Hui, M., Aneja, S.: Deep learning algorithm to predict pathologic complete response to neoadjuvant chemotherapy for breast cancer prior to treatment. J. Clin. Oncol. 40(16_suppl) (2022). https://doi.org/10.1200/JCO.2022.40.16 3. Meti, N.: Machine learning frameworks to predict neoadjuvant chemotherapy response in breast cancer using clinical and pathological features. JCO Clin. Cancer Inf. 5, 66–80 (2021). https:// doi.org/10.1200/CCI.20.00078 4. Park, H.-S., et al.: Machine learning models that integrate tumor texture and perfusion characteristics using low-dose breast computed tomography are promising for predicting histological biomarkers and treatment failure in breast cancer patients (2021). https://doi.org/10.3390/can cers13236013
Breast Cancer Progression Prediction
7
5. Eun, N.L.: Texture analysis with 3.0-T MRI for association of response to neoadjuvant chemotherapy in breast cancer. Radiology 294(1), 31–41 (2019) 6. https://www.fmcgastro.org/textes-postus/no-postu_year/imagerie-en-cancerologie-au-delades-criteres-recist/#:~:text=Les%20crit%C3%A8res%20RECIST%2C%20d%C3%A9v elopp%C3%A9s%20pour,communication%20avec%20les%20m%C3%A9decins%20r% C3%A9f%C3%A9rents
Study of the Germination of Wild and Cultivated Blackberries of the Northern Region of Morocco Amal Maurady1,2(B) , Malika M’guil3 , Dounia Harama1,4 , Iman Touati1 , Naima Bel Mokhtar1 , Soumaya El Ismaili1 , Leila Karimi2,4 , Mohammed Reda Britel1 , and Ahlam Hamim4 1 Laboratory of Innovative Technologies, National School of Applied Science, Abdelmalek
Essaadi University, Tetouan, Morocco [email protected] 2 Faculty of Sciences and Technics of Tangier, Abdelmalek Essaadi University, Tetouan, Morocco 3 Fundamental Sciences Department, Faculty of dental medicine, Mohammed V University in Rabat, Rabat, Morocco 4 National Institute for Agricultural Research, Tangier, Morocco
Abstract. Rubus species germinate slowly due to the dormancy, that is, the nongermination of mature, intact, and viable seeds, despite the necessary conditions. For this reason we investigate the effect of scarification by sulfuric acid combined with boiling water, and gibberellic acid alone at two concentrations of 125 and 200 mg/l respectively in wild and cultivated blackberries of the north of Morocco in order to remove dormancy and increase the germination of berries. Different hormones combinations of gibberellic acid (GA3), potassium nitrate, and kenitin was tested in light conditions at 24 °C in cultivated varieties of blackberries Tupy and Prime-Ark 45, and the variety of the wild blackberry (Rubus fruticosus) grown in Slokiya region in the north of Morocco. The method of scarification by H2 SO4 for 2 h followed by a combination treatment of two hormones GA3 (2.04 mg/l) and KNO3 (34 mg/l) gave us the best germination percentage of 40%. The treatment of GA3 at the concentration of 125 mg/l during 48 h of wild blackberries soaking gives a better germination rate for the wild blackberry in comparison to the cultivated blackberries. Chemical scarification by sulfuric acid combined with boiling water indicated that the best germination rate was recorded in the wild variety followed by cultivated blackberry, and this was performed during 30 min of soaking in sulfuric acid at a concentration of 96% while combining it with boiling water has no effect on the germination variation of these three varieties. Keywords: Scarification · in vitro culture · blackberries · germination · sulfuric acid · hormones · light
1 Introduction The genus Rubus is a widely diversified taxon that includes more than 750 species (Thompson 1995, Strick et al. 2007) in 12 subgenera and occurs on all continents except © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 8–18, 2023. https://doi.org/10.1007/978-3-031-35248-5_2
Study of the Germination of Wild and Cultivated Blackberries
9
Antarctica (Hummer 1996). Berries have now become a common fruit in trade opportunities, particularly in North America and the European Union. Berries have gained attraction due to a combination of factors, including improved cultivars, increased marketing efforts and fruit availability, and an overall increase in the consumption of berries, especially in the form of fresh fruit, in many parts of the world. It is estimated that the blackberries cultivated exceed 25,000 ha worldwide (Strik et al. 2008). Blackberry is one of the most commercially valuable fruits in the world and the most cultivated in regions ranging from 1,200 to 3,000 m above sea level (D’Agostino et al. 2015). Highly perishable fruit, rich in vitamin C and rich in water, native to the American tropical highlands, mainly Colombia, Ecuador, Panama, Guatemala, Honduras, Mexico, and El Salvador. The most studied blackberries are the economically important blackberries (subg. Rubus), they have a deep double dormancy caused by a hard integument and one or more additional mechanisms such as an impermeable integument (exogenous dormancy), chemical inhibitors or the presence of a dormant embryo (endogenous dormancy) (Taylor 2005; Zasada et al. 2003: Cain et al. 2003). Growth-inhibiting substances are concentrated in the endosperm and the testa in dormant blackberry seeds ‘Lawton’. External dormancy is determined by the outer layer of the seed (endocarp); it limits the absorption of water and oxygen and mechanically prevents the embryo from swelling. This dormancy is interrupted if the seed shell is removed, while internal dormancy is regulated by biochemical or biophysical processes that occur during post-maturation at 0–5 °C (AbdAlla et al. 2015; Taylor 2005, Diez et al. 2013). Several studies have been performed as part of removing and reducing the dormancy of berries, including a study that was carried out on the seeds of mulberries of 17 species of Rubus in three subgenera, in order to improve seed scarification and germination protocols, they found that germination after scarification by sulfuric acid was significantly better than Nao Cl, as well as seeds treated with GA3 + KNO3 or smoke germinate significantly better than other treatments; Georgia 3 + KNO 3, (Wada 2009). A similar study was done in Colombia on purple passion fruits evaluated different treatments to improve the pre-germination of purple passion fruit seeds (control, apical and basal seed cuts, temperature alternation, photoperiod, application of gibberellic acid and immersion in 96% of sulfuric acid) and determine the mycorrhizal dependence of this species to the AMF, they found that the treatment with the most significant effect in reducing the dormancy of purple passion fruit seeds is immersion in 96% sulfuric acid for 20 min, and that this species is highly mycorrhizal dependent, when combined with 0.02 mg/l of P in the soil solution (Gil et al. 2015). Another study carried out as part of characterizing seed dormancy in blackberry (R. glaucus) and associated taxa, and to develop release protocols, in order to facilitate the collection and preservation, as well as the subsequent development, of improved cultivars, proved that despite the viability of the seeds (tetrazolium test) there is the impermeability of the integument that prevents imbibition, which is considered an expression of exogenous dormancy; however, this was inverted by immersing the seeds in 5.25% sodium hypochlorite for 16 to 21 h, and that the seeds germinated in light and dark conditions. Thus, the highest germination figures were obtained with seeds harvested
10
A. Maurady et al.
during the dry season incubated in the dark and with rainy season seeds incubated in light conditions (Diez et al. 2013). The objective of this work is to study the effect of hormonal treatment by the gibberellic acid alone at two concentrations (125 and 200 mg/l), different hormones combinations (gibberellic acid, potassium nitrate, and kenitine), and sulfuric acid combined with boiling water on the germination and growth of varieties of the crop blackberry (Tupy and PrimeArk 45) grown in Larache, and wild blackberry (Rubus fruticosus) from Slokiya, and validate its results by a statistical study using the R software to process the data, then the comparison of our results with the other research that has been carried out in this context.
2 Methods 2.1 Seed Materials The wild blackberry (Rubus fruticosus) (slokiya) were collected in the month of July 2020 in the Tangier region in Morocco (35°47’16.3”N 5°54’31.3”W), and cultivated blackberry fruit (Tupy and PrimArk 45) were provided by National Institute of Agricultural Research of Tangier. The seeds of the three blackberries varieties were analysed by germination tests. 2.2 Seed Preparation The seed fruits of cultivated blackberries varieties and wild blackberry were gently crushed in a mortar, then squeezed into the water in a beaker, placed on joseph paper, and air-dried overnight. The seeds were stored in a sachet and placed in a desiccator. The seeds of the two varieties (Tupy and Prim-Ark 45) of cultivated blackberry and wild blackberry (slokiya) were dried for a week in an oven at a temperature of 30 °C. The dry weight of the seeds was thus measured. 2.3 Preparation of Hormones and Sterilization of Equipment We used 3 hormones gibberellic acids (GA3), kinetin (C10 H9 N5 O), and potassium nitrate (KNO3 ), and all possible combinations of these 3 hormones (GA3 + KNO3, GA3 + kinetin, KNO3 + kinetin, and GA3 + KNO3 + Kinetin). 2.4 Germination Experiments 2.4.1 Treatment of Seeds with Sulfuric Acid Under a vertical laminar flow hood, the seeds of both varieties of blackberries tree Grown in Larache (Tupy and Prim-Ark 45) and of wild blackberry have been set up in boxes of Petri dish then we have added 5ml of sulfuric acid at a concentration of 96% for, 1 h 20 min, 2 h and 3 h (Wadaa and Reed, 2011). Then the treated seeds were rinsed 3 times with distilled water for 5 min. The seeds, thus prepared, are deposited on sterile filter paper in Petri dishes and separated using a lanceolate needle.
Study of the Germination of Wild and Cultivated Blackberries
11
2.4.2 Treatment with Sulfuric Acid (H2 SO4 ) Combined with Boiling Water Under a vertical laminar flow hood, the seeds of both varieties (Tupy and Prim-Ark 45) and of wild blackberry were set up in petri dishes, then we added boiling water for 3 min then 5 min (A). Then the seeds were soaked in 96% sulfuric acid with the following durations (B): • • • •
Boiling water at 100 °C for 3 min + H2 SO4 Boiling water at 100 °C for 5 min + H2 SO4 Boiling water at 100 °C for 3 min + H2 SO4 Boiling water at 100 °C for 5 min + H2 SO4
for 15 min for 15 min for 30 min for 30 min
After soaking, the treated seeds were disinfected with bleach (NaClO) at 5% for 5 min, then rinsed 3 times with distilled water, for each rinse a slight stirring was performed. The seeds are placed on sterile filter paper in Petri dishes (20 seeds per box) and separated using a lanceolate needle, then soaked with 3 ml of distilled water (C). The test was performed in light conditions, with a repetition of each soak and the control. In total, 10 Petri dishes are prepared for treatment with H2 SO4 in each variety. The control seeds were germinated only with distilled water without the H2 SO4 treatment. The Petri dishes were hermetically sealed with parafilm and were placed in a germination chamber at 24 ± 1ºC, and 60% relative humidity. The Petri dishes incubated in the oven were covered with aluminum foil to prevent the entry of light (D). The evolution of germination was observed every 48 h for 30 days. 2.4.3 Treatment with Gibberellic Acid GA3 Under a vertical laminar flow hood, the seeds of two varieties of blackberries grown locally in the greenhouses (Tupy and Prim-Ark 45) and of wild blackberries were set up in Petri dishes, then we added the bleach (NaClO) at 5% for 5 min and then rinsed 3 times by distilled water (A). Then the seeds were soaked in gibberellic acid with the following concentrations and durations (B): • • • •
Soaking in GA3 at 125 mg/l for 24 h Soaking in GA3 at 125 mg/l for 48 h Soaking in GA3 at 200 mg/l for 24 h Soaking in GA3 at 200 mg/l for 48 h
After soaking the treated seeds were rinsed 3 times with sterile distilled water, for each rinse a slight stirring was performed. The seeds are placed on sterile filter paper in Petri dishes (20 seeds per box) and separated using a lanceolate needle, then soaked with 3 ml of sterile distilled water (C). The test was performed under light, with a repetition of each soak and the control. In total, 10 Petri dishes are prepared for treatment with GA3 in each variety. The control was germinated only with distilled water without the GA3 treatment. The Petri dishes were hermetically sealed with parafilm and were placed in a germination chamber where the temperature is set at 24 ± 1ºC, with a relative humidity of 60%. The Petri dishes were covered with aluminium foil to prevent the entry of light (D). The evolution of germination was monitored every 48 h for 30 days.
12
A. Maurady et al.
2.5 Germination Treatment 5 mm Petri dishes were used; a filter paper is placed in each petri dish. The test was performed in light conditions, with 3 repetitions of each hormone and the control. In total, we prepared 48 Petri dishes for treatment with sulfuric acid for 1 h 20 min, 48 Petri dishes for the duration of 2 h, and 48 boxes for treatment with sulfuric acid for 3 h which is 144 in total for each variety. The control seeds were scarified and germinated only with distilled water. The treatments of scarified seeds with H2SO4 for 1 h 20 min, 2 h and 3 h were as follows: (1) sterile distilled water, (2) gibberellic acid (250 mg/l GA3) (koyuncu et al. 2005) (3) potassium nitrate (100 mg/l) (4) Kinetin (0.05 mg/l), (5) 2.04 mg/l GA3 + 34 mg/l KNO3 (Wadaa and Reed, 2011) (6) 2.04GA3 mg/l (Wadaa and Reed, 2011) + 0.005 mg/l Kinetin, (7) 15 mg/l KNO3 + 0.005 mg/l of kinetin, (8) 0.1 mg /l of GA3 + 8 mg/l of KNO3 + 0.0005 mg/l of Kinetin. All treatments were replicated with 10 seeds per petri dish (5 mm) and 3 repetitions in light (n = 30. In all Petri dishes, a filter paper is deposited, then the seeds have been deposited (10 seeds per box), the Petri dishes have been soaked with 4 ml of sterile distilled water for control and 4 ml of the different hormones and hormone combinations. The Petri dishes were hermetically sealed with parafilm and were placed in a germination chamber in conditions where the temperature is set at 24 ± 1ºC and relative humidity of 60%. The Petri dishes were covered with aluminum foil to prevent the entry of light. Every 3 days, a check was performed, and the germination rate was noted in light conditions for each duration of AS and each hormone combination. %germination = number of sprouted seeds/Total number ∗ 100 A seed that has a radicle of 2 mm is considered germinated (Koyuncu et al. 2000). 2.6 Data Treatment The results were expressed by the calculation of the averages of the observed results. They are treated by the test of variance (ANOVA) (Girden, 1992). The analysis is performed using R-type software (R Core Team 2022). The aim of this study is removing and reducing the dormancy of the berries, in order to improve seed scarification and germination protocols for the various species of blackberries often used for the selection of improved cultivars. They found that germination after scarification by sulfuric acid was significantly better than Nao Cl, as well as seeds treated with GA3 + KNO3 or smoke germinate significantly better than other treatments; Georgia 3 + KNO3 , (Wada 2009).
3 Results 3.1 Imbibition Test According to the curve, we notice that the primeArk45 variety has an exponential and regular water absorption capacity for 8 h, compared to the other two Slokiya and tupy varieties which reach their best weight at 7 h and then begins to decrease (Fig. 1).
Study of the Germination of Wild and Cultivated Blackberries
13
Fig. 1. Variation in seeds weight of the three blackberries varieties with imbibition time
The ANOVA test showed that the weight significantly depends on the interaction between the seeds of variety and the imbibition time because the p-value = 1.84e-12 is less than 0.05. The imbibition time affects the weight differently depending on the varieties seeds used, and no significant difference was observed between the imbibition time from 2 h to 7 h for the two varieties Slokiya and Tupy. However, for the variety PrimeAKr 45, significant differences were observed. Effectively the weight increases significantly with the imbibition time. In our study the imbibition test was used to detect the absence or presence of physical exogenous dormancy in the three varieties, the imbibition results for 8 h showed us that there is a significant difference in seed weight change due to water absorption (p = 1.84e−12 < 0.05) compared to time zero for the three seed types, this indicates the permeability of the seed coat. The absence of a significant difference between the imbibition time from 1h to 8h for the two varieties Slokiya and Tupy, and its presence for the variety PrimeAKr45, the weight increases significantly with imbibition time, can be explained by a difference in water absorption capacity in the three soaking varieties due to a slight exogenous dormancy in slokiya and tupy and not the variety Prime-AKr 45. Our results are not in accordance to the results of Díaz Diez (2013) and Wada (2009) who recorded a lack of imbibition in several Rubus spp. For this reason, several other studies have emphasized the importance of scarification (Raymond et al. 2003) (Gauthier et al. 2016).
14
A. Maurady et al.
3.2 Gibberellic Acid Treatment According to the results obtained, the best seeds germination rate during 48 h of soaking in GA3 at a concentration of 125 mg/l of Tupy variety was best one followed by the variety Prime-AKr45 and Slokiya respectively. For soaking in GA3 at a concentration of 200 mg/l, the best germination rate was observed for Prime-AKr45 and tupy seeds during 24 h and for the wild blackberries 48h gives better results (Fig. 2).
Fig. 2. Variation in the germination rate according to the hormone and the application time for each variety. Error bars indicate SE for (number replicates = 2) of (number seeds = 10) and p-value = 0.00157
According to the results of the GA3 treatment obtained, the best germination rate was in favor of the cultivated blackberries, a rate of (53% ± 2.5) was that of the Tupy variety during 48 h of soaking in GA3 at a concentration of 125 ppm, followed by (43% ± 12.5) and (38% ± 2.5) during 24 h of soaking in GA3 at a concentration of 200 mg/l and for 48 h of soaking in GA3 at a concentration of 125 mg/l respectively for the PrimeArk45 variety. For the wild variety, the best germination rate was detected during 48 h of soaking in GA3 at a concentration of 125 mg/l for a value of (20% ± 2.5) of germination. The ANOVA test showed that the application of different concentrations of GA3 and varieties significantly affects the germination rate (p = 0.001). In addition, the effect of concentrations significantly decreases in application time and cultivar used (Interaction Concentration_AG vs Time: p = 0.001 and Variety vs AG Concentration_AG Interaction: p = 0.023). For the variation of concentration of GA3, the test recorded that the GA3 concentration significantly affects the germination rate for the Slokiya and Tupy varieties with a p-value = 0.040 and a p-value = 0.035 respectively less than 0.005.
Study of the Germination of Wild and Cultivated Blackberries
15
Statistical analysis showed that the application of different concentrations of gibberellic acid significantly affects the germination rate, hence the 3 varieties experienced a significant response with a concentration of 125 mg/l and a response of only 200 mg/l for the prime-Ark45 variety, so it is assumed that the concentration of 200 ppm can have an impact on the viability of Slokiya and Tupy. In addition, the effect of GA3 concentrations significantly decreases the application time (Interaction GA concentration vs time: p = 0.001), hence the 3 varieties experienced a significant response to an application time of 48h with a concentration of 125ppm followed by a response to an application time of 24h with the same concentration. Our results are consistent with that of Park et al. (2015) who found that gibberellic acid exerted a spectacular action on the elongation of peanut stems and twigs. Wada and Reed (2011) observed that the application of GA3 alone on various species of raspberries and blackberries gives better germination than NaOCl for four of the six species. 3.3 Treatment of Rubus Seeds with Sulfuric Acid (H2 SO4 ) Combined with Boiling Water According to the results obtained, the best germination rate was recorded in the slokiya variety 68% (±2.5) during 3 min of soaking in boiling water and 30 min of soaking in H2 SO4 , followed by the 55% (±2.5) of germination for 5 min of soaking in boiling water and 30 min of soaking in H2 SO4 for the same cultivar, and 43% (±2.5) for 3 min soaking in boiling water and 30 min soaking in H2 SO4 for the Tupy variety. Thus, a germination rate of 38% (±2.5) during 3 min of soaking in boiling water and 30 min of soaking in H2 SO4 for the PrimeArk45 variety. The variance calculated according to the time of boiling water, the time of application of Hormone_H2 SO4 and the blackberries variety. The duration of H2 SO4 hormone application and the variety significantly affected the germination rate (p < 0.05), while no effect was observed by combining it with boiling water. The result of the ANOVA test is described in Table 1. Table 1. The effect of application of Hormone_H2 SO4 and boiling water at different times for the three Effect on the rate of germination of blackberries varieties
DL Significance level (p value)
Varieties
2
0.00641
Soaking time of boiling water
1
0.27069
Soaking Times of H2 SO4
1
0.00164
Varieties vs Soaking time of boiling water
2
0.73375
Varieties vs Soaking Times of H2 SO4
2
0.43984
Soaking Times _H2 SO4 vs Soaking time of boiling water
1
0.08513
Soaking Times of H2 SO4 vs Soaking time of boiling water 2 vs Varieties
0.41661
DL: represents the degree of freedom
16
A. Maurady et al.
Furthermore, the variety significantly affects the germination rate and analyzed with pairs of different blackberries varieties groups with the Tukey Honest Significant Differences test (Tukey HSD) in Table 2. Table 2. Results of the Tukey test HSD Pairs of blackberries varieties
Significance level (p value)
Slokiya vs Prime-Ark45
0.0124216
Tupy vs Prime-Ark45
1.0000000
Tupy vs Slokiya
0.0124216
Fig. 3. Variation in germination rate according to H2 SO4 application time for each variety. Error bars indicate SE for (number replicates = 2) of (number seeds = 10) and p-value = 0.00164
The ANOVA test showed that the duration of application of H2 SO4 and the type of cultivar significantly affected the germination rate (p < 0.05), while no effect was observed by boiling water. In these datasets, the cultivar significantly affects the germination rate, so that we know which pairs of groups are different we applied the Tukey HSD test (Tukey Honest Significant Differences). From the Tukey HSD test, we notice that the best germination was in favor of pairs of slokiya groups (wild variety) compared to primeArk45 and tupy (p = 0.012), while no significant difference was detected for cultivated blackberry trees (Fig. 3). In order to optimize germination in the three varieties, we returned to other tests, including treatment of Rubus blackberry seeds with gibberellic acid (GA3), and chemical scarification with sulfuric acid (H2 SO4 ) combined with boiling water.
Study of the Germination of Wild and Cultivated Blackberries
17
For chemical scarification by sulfuric acid combined with boiling water, the results obtained reported that the best germination rate was recorded in the wild variety, followed by cultivated blackberries. The duration of application of H2 SO4 and the cultivar significantly affect the germination rate (p < 0.05), from which it is noted that the best germination rates were reported during 30 min of soaking in H2 SO4 at a concentration of 96%, while boiling water had no effect on the germination rate was observed. The Tukey test showed a difference in germination between wild blackberry and cultivated blackberry (p = 0.012) in favor of wild blackberry, while cultivated blackberry varieties represent no significant difference between them. These results are consistent with the results of Gil et al. (2015), that indicated that treating the seeds with 96% sulfuric acid for 20 min improves the germination process of the purple passion fruit without compromising its viability.
4 Conclusion In the light of this study, we can deduce that the best treatment for improving the germination of the wild variety is the combination between soaking in boiling water and chemical scarification with sulfuric acid at a concentration of 96% was for 30 min that induce an increasing the germination rate to 68% while boiling water has no effect. This result demonstrates the existence of physical dormancy related to the seed Testa. Hormonal treatment with GA3 showed that the best germination rate was recorded in the cultivated blackberry variety at a concentration of 125 mg/l for 48 h imbibition. Furthermore, the 200 mg/l of GA3 concentration can affect the viability of the Slokiya and Tupy varieties.
References AbdAlla, M.M., Mostafa, R.A.A.: In Vitro Propagation of Blackberry (Rubusfruticosus L.). Assiut J. Agric. Sci 3, 46 (2015) Acosta, O., et al.: Potential of ultrafiltration for separation and purification of ellagitannins in blackberry (Rubus adenotrichus Schltdl.) juice. Sep. Purif. Technol. 125, 120–125 (2014). https://doi.org/10.1016/j.seppur.2014.01.037 Cain, M.D., Shelton, M.G.: Fire effects on germination of seeds from Rhus and Rubus: competitors to pine during natural regeneration. New For. 26(1), 51–64 (2003). https://doi.org/10.1023/A: 1024406209842 Diez, D., Arturo, C., et al.: Dormancy and germination of castilla blackberry seeds (rubus glaucus benth)/latencia y germinación de semillas de mora de castilla (rubus glaucus benth). Revista Facultad Nacional de Agronomía Medellín (2013) D’Agostino, M.F., et al.: Optimization of a solid-phase microextraction method for the gas chromatography–mass spectrometry analysis of blackberry (Rubus ulmifolius Schott) fruit volatiles. Food Chem. 178, 10–17 (2015). https://doi.org/10.1016/j.foodchem.2015.01.010 Fira, A., Clapa, D., Plopa, C.: New aspects regarding the micropropagation of blackberry cultivar ‘Thornless evergreen.’ Bull. UASVM Horticulture 67(1), 106–114 (2010) Gauthier, M.M., Lambert, M.C., Bédard, S.: Effects of harvest gap size, soil scarification, and vegetation control on regeneration dynamics in sugar maple-yellow birch stands. For. Sci. 62(2), 237–246 (2016)
18
A. Maurady et al.
Girden, E.R.: ANOVA: Repeated measures. Sage (1992) Koyuncu, F., Sesli, Y.: Effect of different stratification periods and soaking times in water on the germination and seedling growth of walnut. In: Proceedings of the 11 National Nursery Symposium, Izmir (2000) Koyuncu, F.: Breaking seed dormancy in black mulberry (Morus nigra L.) by cold stratification and exogenous application of gibberellic acid. Acta Biol. Cracov. Bot. 47(2), 23–26 (2005) Park, E.-J., et al.: Establishment of seed treatment for healthy production of peanut sprout. J. Environ. Sci. Int. 24(6), 755–762 (2015). https://doi.org/10.5322/JESI.2015.24.6.755 Ramírez, G., Joaquín, G., et al.: Germination and growth of purple passion fruit seedlings under pregermination treatments and mycorrhizal inoculation. Pesquisa Agropecuaria Tropical 45(3), 257–265 (2015) Raymond, J., et al.: Long-term angiographic recurrences after selective endovascular treatment of aneurysms with detachable coils. Stroke 34(6), 1398–1403 (2003). https://doi.org/10.1161/ 01.STR.0000073841.88563.E9 R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria (2022). https://www.R-project.org/ Strik, B.C., Clark, J.R., Finn, C.E., Banados, M.P.: Worldwide production of blackberries (2008) Taylor, K.: Biological flora of the British isles: rubus vestitus Weihe. J. Ecol. 93(6), 1249–1262 (2005). https://doi.org/10.1111/j.1365-2745.2005.01076.x Thompson, E., Strik, B.C., Clark, J.R., Finn, C.E.: Flowering and fruiting patterns of primocanefruiting blackberries. HortScience, 42(5), 1174-1176 (2007) Wada, S., Reed, B.M.: Standardizing germination protocols for diverse raspberry and blackberry species. Sci. Hortic. 132, 42–49 (2011). https://doi.org/10.1016/j.scienta.2011.10.002 Wada, S.: Evaluation of Rubus seed characteristics: seed coat morphology, anatomy, germination requirements and dormancy breaking. Oregon State University, 2009 Zasada, J.C., John, C.T., III, Rubus, L.: The Woody Plant Seed Manual. USDA Forest Service, pp. 1629–1638 (2003)
The Impact of Covid 19 on Recommendation Platforms Amina Samih(B) , Abderrahim Ghadi, and Abdelhadi Fennan Affiliation Faculty of Sciences and Techniques, DIS Laboratory University Abdelmalek Esaadi, Tangier, Morocco [email protected]
Abstract. Recommender Systems that use artificial intelligence algorithms have achieved much lately, prompting exact expectations in numerous e-commerce areas. Nonetheless, this climate experienced unexpected changes with the beginning of the corona pandemic fixated on a remarkable expansion in the volume of clients and quick adjustments in client practices and profiles. This paper examines the effect of the corona pandemic on the recommender systems scene and spotlights new and abnormal customers. We detail how multicriteria, psychological, explainable methods, and online artificial intelligence can recognize and adjust to changes in purchaser practices and utilize profiles to give exact and convenient forecasts. Keywords: Recommender systems · Covid 19 · Cold start · Grey sheep · Machine learning · Multicriteria methods
1 Introduction With the advent of Recommender systems [1], which are behind many applications, social networks, shopping sites, and more, we have indeed already come across the following phrase "users who liked this article also liked this one". This is what recommendation systems are all about, and their use is growing among big names like Netflix, Amazon, Pinterest, and Facebook. A recommendation system [2] filters data, reusing it to suggest new data that may interest the user. These data can be diverse: films, series, music, articles, etc. The development of such systems has enabled a revolution in the user experience thanks to a range of services adapted to each profile and better communication action thanks to targeted advertising [3]. The right products are offered to the right people, who are more likely to buy them [4]. What impacts the e-commerce sector. The trend of e-commerce has grown enormously during this health crisis linked to the pandemic of the new coronavirus. This strong enthusiasm has been observed in practically all sectors. It has prompted salespeople to review their communication strategies to adapt to this new reality and maintain customer ties [5]. Thus, the commercial sites recorded significant success in terms of sales during this period. Moreover, no one can doubt that e-commerce today is intended to be a real lever for performance and © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 19–34, 2023. https://doi.org/10.1007/978-3-031-35248-5_3
20
A. Samih et al.
growth [6], making it possible to compensate for the closure of physical points of sale and avoid any risk of contamination by the virus. Indeed, new consumers who may be cold-start users [7] (use the system for the first time) or grey sheep users [8] (have preferences out of the ordinary) did not use the internet for their daily purchases, took on new habits during the confinement that could last even after this period. What raises the urgency to propose algorithms of recommendations to retain the new customers will be a boom in research. Recommendation systems are widely used applications of machine learning technologies; For instance, [9] suggests using machine learning to learn patterns that link user profile information to user interests. This technique can determine if any user profile variable(s) and a user interest have a relationship. It can also select the best threshold for the user profile information to address cold start users’ problems; recommendation systems have widely served the international e-commerce giants [10]. We can quote Amazon, which uses recommendation algorithms to personalize the online store for each customer [11]. The store changes dramatically based on customer interests, showing programming products to a software engineer or baby toys to young parents. Indeed, we can also mention the Netflix platform [12], which recommends movies or the content consumed previously that will influence those offered later. Netflix honestly conducts behavioral analysis through machine learning, and the algorithm reviews several data. Nonetheless, the COVID-19 epidemic impact dramatically the e-commerce landscape shifted. This position paper offers some observations on the field’s current state and some suggestions for the future. The rest of this paper is organized as follows. Section 2 introduces the basic concepts., In Sect. 3, we discuss the current impact of COVID-19. In Sect. 4, we focus on the abovementioned challenges that Recommendation Systems are currently facing, in Sect. 5 we present our proposed solution. we highlight future research directions and we conclude the paper in Sect. 6.
2 Background The primary purpose of this section is to explore the background of recommender systems and, respectively, cold start and grey shep users. 2.1 Recommender Systems First of all, a recommendation system is a business tool; it boosts a company’s revenue by up to 30%; Today, users do not want products offered on the internet that they have already purchased or that are not of interest to them. For this, recommendation systems aim to understand users’ behavior; To make their life easier, then the site or the application will gain their trust [13].
The Impact of Covid 19 on Recommendation Platforms
21
Recommender systems slip more or less discreetly into our digital lives [14] and punctuate our daily lives when browsing the web. We are unaware of this, yet recommendation is ubiquitous in all major digital industries, such as e-commerce, online press, videos, and music streaming services. Of course, social networks also use it extensively. With the advent of information overload, consumers have neither the time nor the patience to explore these suggestions. Therefore, the main idea behind recommender systems is to address the problem mentioned above and aid users in narrowing down their list of choices by presenting and treating data via machine learning algorithms [4]. That appeared at the beginning of year 1950s [15]as a subfield of artificial intelligence. In general, the goal of machine learning is to understand the structure of the data and integrate it into models that can be understood and used by everyone. This learning involves looking for an underlying set of forms (or models) helpful in understanding relationships in data that might not be exactly similar to those on which learning occurred.[16]. Based on the user’s preferences, all recommendation platforms are powered by machine learning to suggest which movies or TV shows to watch…. The domain of machine learning is still in development; there are mainly three types of learning (Fig. 1):
Fig. 1. The main models of machine learning
• Supervised machine learning • Unsupervised machine learning • Machine learning by reinforcement. Recommender systems use three principal approaches [17]: content-based filtering, Collaborative filtering, and hybrid recommendation (see Fig. 2).
22
A. Samih et al.
Fig. 2. Recommendation platform
The first one analyzes the attributes of items to generate predictions. Recommendations are made based on features extracted from the content of items with which a user has interacted in the past. Articles that are primarily related to top-rated articles are recommended to the user [18]. For example, if I listen to a song in a particular genre, I receive recommended songs in that specific genre. Piece attributes like title, year of release, singer, and instruments are also helpful in identifying similar music. Content-based filtering techniques do not need other users’ profiles because they do not influence recommendations. So, if the user’s profile changes, this technique can still adjust the recommendations within a short period [19]. Content-based filtering can also recommend new items even if users do not provide ratings or reviews. So even if the database does not contain user preferences, the accuracy of recommendations is not affected. That said, effective content-based filtering requires detailed descriptions of items [1]. The second approach works by creating a preference database for items from previous users [20]. It then matches new users with relevant interests and preferences by calculating the similarities between their profiles and previous ones to make recommendations. A user receives recommendations for items they have not rated before but have already been rated positively by other users with similar interests [21]. Amazon is a typical example of collaborative filtering. They use scalable, element-to-element collaborative filtering techniques to produce high-quality, real-time recommendations. Amazon [10] makes it easy for customers to compare similar items on product detail pages, create urgent product recommendations, suggest products on category pages, and more.
The Impact of Covid 19 on Recommendation Platforms
23
The third approach combines different recommendation techniques to achieve better system optimization to avoid the limitations and problems of pure recommender systems [22]. Research has shown that such a combination of algorithms provides more accurate and efficient recommendations than a single algorithm because another algorithm can overcome the drawbacks of one algorithm [2]. Aside from the above approaches, there are other ways to group recommender systems, such as demographic filtering, which examines demographic data like age, gender, education, address, and other factors to determine user commonalities [49]. Moreover, there is Knowledge-based Recommendation, which is based on recommendations that have previously been stored in databases or knowledge bases and so are not influenced by recent ratings or preferences [48]. The quality evaluation of these handy systems is one of the most critical challenge in the recommendation sector. In general, these systems are assessed in various ways, many of which are incomparable. Indeed, as we said, recommender platforms are among the most successful uses of machine learning techniques [4], and each algorithmic approach has proponents who say it is superior for a particular goal. As a result, determining the optimum algorithm is inherently tricky. Algorithm correctness, diversity and innovation, coverage, and uncertainty have all been proposed as metrics for judging the success of a recommendation approach [3]. However, quality is not always connected with systemcentric criteria, and correct recommendations may not always be the most valuable to users. As a result, assessment metrics should be more user-centric, considering elements that influence user happiness and drive them to act on the advice (purchase, listen, watch…). In this regard, a good recommender system, in our opinion, should be transparent able to explain (and hence persuade) to the active user why certain goods are recommended. An excellent yet complex search topic is moving beyond accuracy measurements to create and evaluate an explainable recommendation platform. Making a recommender system result interpretable is challenging because most of these systems are based on machine learning algorithms that are "not transparent" by nature. Fortunately, XAI [1], a new field of study, proposes ways for making artificial intelligence systems results in more human-friendly. The recommendation is divided into three phases: information gathering, learning, and producing recommendations [23]. The goal of information collecting is to learn more about the users. The quality of recommendation is highly tied to the quality of information about the users in the system, as numerous works have pointed out. The system receives this data in the form of user feedback. Three forms of feedback could be present in a recommender system: • Explicit feedback: The user provides a rating to products through the system interface. • Implicit feedback, in which the system utilizes user behaviors, history, and purchases. • Hybrid feedback combines explicit and implicit feedback. The learning phase starts when an algorithm is launched to learn users’ feedback. Finally, the system can produce recommendations in two forms, generating predictions in scores, which present the percentage that user Ux will like an item Ix, or listing the top N items that maybe interest a particular user.
24
A. Samih et al.
As previously stated, recommender systems have shown to be quite effective in tracking customers with similar profiles. Nonetheless, users’ preferences are unknown when they are using, for example, e-commerce systems for the first time, as was the case with the increase in online buying during the corona pandemic. Furthermore, a growing number of users have unusual or exotic tastes, making it increasingly difficult for the system to match their preferences with the current customers’ data. These users may decrease the accuracy of recommendations, resulting in incorrect or illogical recommended items. To this end, we propose to explore these particular types of users in the next sub-sections. 2.2 Cold Start Users A cold start user is a situation where the recommender system does not have sufficient information about a user to make meaningful predictions. This is one of the significant problems that reduce the system’s performance. The profile of this new user will be empty since he has not rated an article, so its taste is not known. A user must evaluate a certain number of items before the system interprets his preferences and provides relevant recommendations [24]. This problem is known in the literature as the user cold-start problem. Users’ tastes, historical data about what they like and dislike, and item ratings and reviews are utilized in recommendation systems to match them with other users. All those information are not available in cold-start conditions, making it impossible for the system to determine similarity ratings [25]. Indeed, during the present corona epidemic, the massive increase in the usage of e-commerce websites has underlined the need of, and difficulty in, delivering reliable suggestions to many new users [26]. We can use the collaborative filtering-based recommender system to tackle this challenge, using item descriptions to find the corresponding ones [27]. Other systems merely show these consumers a list of pre-made suggestions. However, these methods may work for some consumers; they frequently rely on the presentation of redundant lists, which causes these users to lose trust. [28] proposed to employ intimate machine learning techniques, in which a cold start user is given a list of questions to answer to create a first preference profile. Because of the time it takes to develop a profile; privacy issues may also drive them away. Recently, we proposed [1] a hybrid recommendation framework that uses both Knn and word2vec (graph embeddings algorithm) models to make recommendations for new users. 2.3 Grey Shep Users Grey shep users are users with different, out-of-the-ordinary tastes won’t have many similar users. Therefore, it will not be easy to make relevant recommendations for this type of user. These users are difficult to detect and categorize; they are challenging to detect and classify [29]. These users give their feedback to some extent. On the other hand, their choices rarely match the bulk of the system’s preferences. Unlike cold starts, the system may already contain the data required to calculate similitude and generate recommendation results due to their distinct tastes and qualities.
The Impact of Covid 19 on Recommendation Platforms
25
By utilizing the user’s profile and the contents of the items being recommended, pure collaborative filtering can fix this issue where items are suggested. Collaborative filtering can also resolve sparse ratings [2]. Combining content and collaborative filtering approaches may result in more unexpected and novel recommendations [18]. With offline clustering techniques such as k-mean clustering, grey sheep users can be recognized and separated from others. As a result, performance improves, and suggestion inaccuracy is reduced. By reusing outlier recognition techniques based on the distribution of useruser similitudes, [30] suggest a novel strategy for identifying Grey Sheep users. [31] proposed the grey-sheep one-class recommendation (GSOR) framework, which is a framework for creating accurate prediction models that consider both regular and greysheep users; authors demonstrated using one-class classification learning in grey-sheep recommendation systems can improve overall performance and prediction accuracy.
3 Recommender Systems and Covid 19 Given the current context and the unprecedented health crisis, the safest solution is to swap physical activities with online alternatives. We see this, especially in the Retail sector: consumers being invited to stay at home, a significant amount of purchases usually made in stores are, for the moment, made online [32]. Thus, according to a recent study [33], the traffic and the number of conversions on the Supermarkets’ websites have more than doubled (around 2.5 times more) compared to the pre-coronavirus period. Coronavirus has accelerated the expansion of e-commerce to new categories of businesses, consumers, and products. This boom has allowed consumers to access a wide range of products while remaining comfortably and safely at home and businesses to continue operating despite restrictions on physical contact and other containment measures [34]. [33] noted that the health crisis has accelerated the digitization of e-commerce purchases by professionals.34% of B2B e-commerce sites customers consider that the health crisis has led them to develop online orders to the detriment of other channels (e-mail, fax, stores, agencies, telephone, the order was taken by commercial). More than two-thirds believe that this transfer of demands to the e-commerce channel has been achieved. A demand effect explains this acceleration in the development of online "B2B» orders: customers confined to teleworking no longer had the usual systems and equipment to make their purchases. Furthermore, they were therefore encouraged to switch to ecommand platforms. Nevertheless, this is also explained by a large offer. Given the context and perceiving that this acceleration in the digitization of purchases would last, many distributors have refocused or accentuated their efforts on digital. " The experience analytics platform in [33] analyzed 1.8 billion user sessions and 50 million transactions on 1,400 websites worldwide. Between the weeks of February 2020, [35] compared the conversion rate, number of transactions, number of visits, length of sessions to analyze changes in online consumer behavior over the past two weeks. The platform sees a surge in conversions for housewares and groceries and a significant increase in the hours spent on the web searching for essentials. Sales in brick-and-mortar stores rose 16% with consumer fear of a shortage, alongside an 8.1% increase in the industry’s average conversion rate.
26
A. Samih et al.
Online shopping for products in supermarkets increased by a fifth (19.9%), while consumers spent 25.7% more hours searching online. Retailers also saw their sales jump by a fifth (21%), with a solid conversion rate of 14.8%. According to the data, buyers spent 44% more time on these sites. We can say that this health crisis provides an e-commerce opportunity; we can cite French e-commerce; in general, FEVAD [33] (Federation of e-commerce and distance selling) looks back on the impact of Covid 19 on this e-commerce. Indeed, despite the current containment measures, 94% of e-commerce sites are still open in France. Amazon, as an example, had already had the excellent year 2020, driven by the Covid-19 pandemic, which has increased the demand for home delivery around the world tenfold. The year 2021 promises to be just as brilliant for the American e-commerce giant. While doubts remain about its ability to multiply after the pandemic, that is not yet translated into the numbers. The platform’s net profit more than tripled, to $ 8.1 billion for the period from January to March, for a turnover which also vastly exceeded its expectations and those of the market by crossing the $ 100 billion to $ 108.5 billion, up 44% from the same period last year. The present shift in consumer behaviors highlights the necessity of addressing customers’ requests. It also emphasizes the importance of recommending the correct items to the proper clients, such as cold start and grey sheep users, to prevent losing them to competitors and streamline supply chains. The level of online competition is at an all-time high, and a large majority of enterprises must deal with it. Several studies have demonstrated the importance of e-commerce and recommendation Systems (these systems can manage information overload, which is a big challenge today) in all domains during the epidemic. This transition is especially crucial in the healthcare industry, where healthcare professionals have shifted to e-commerce to deliver customized care [36].
4 Cold Start and Grey Shep Users COVID-19 is permanently changing consumer behavior; Consumers are very concerned about the impact of COVID-19, both on health and the economy. Users react differently and have different attitudes, behaviors, and buying habits. All over the world, people are afraid as they struggle to adjust to a new normal. Fear rises as people reflect on what this crisis means for them, but more importantly, what it means for their family, friends, and society in general. Consumers are responding to the situation in various ways. Some people feel anxious and worried, encouraging compulsive purchases of primary and hygiene products. At the other extreme, some consumers remain indifferent to the pandemic and continue to operate as usual, despite recommendations from government and healthcare professionals. Consumer Staples companies will need to understand the reaction of their consumers and develop personalized marketing strategies accordingly. The days of universal marketing are over. The health crisis has led to pressure in the research field of recommender systems. Numerous new customers are joined to existing recommendation platforms as users skyrocket. Otherwise, even for existing Consumers, their priorities have become centered
The Impact of Covid 19 on Recommendation Platforms
27
on the most basic needs, causing the demand for hygiene, cleaning, and essential products to skyrocket as online orders for non-essential product categories are collapsed. So the rating of items stored will not be used (a big challenge). People’s habits are changed since the start of confinement; Instead of rushing out the front door to begin the daily commute to work, the person had the opportunity to relax during the workday before turning on the laptop on the dining room table. So, alarm clock purchases fell 38%, but espresso machine sales rose 12%. Suitcase sales fell 69% - after all, we could not go anywhere. Furthermore, while the local gym was banned, home workout accessories, such as fitness equipment and smartwatches, gained popularity, recording sales growth of 64%. Moreover, dressing to go out is not yet back in fashion: sales of high-heeled shoes and party handbags have fallen by 62% and 56%, respectively. People have "freed themselves from the tyranny of the belt" and have become comfortable with a casual dress which changes the way people buy clothes [37]. All the changes in customer habits have disrupted the algorithms used during the recommendation because the history of users is no longer beneficial because their taste is no longer the same. Another issue is cold-start consumers, as many folks are new to e-transactions. They pose an important gap in the existing history database of users in recommender systems that surge a difficulty to manage good recommendations for them. A significant part of these new users may be grey sheep who would not usually use e-businesses during regular periods. Based on these challenges, the change in preferences creates a data scarcity that is a significant challenge for recommender systems. We cited French e-commerce based on [38] as an example; the first confinement linked to the Covid-19 health crisis largely impacted the e-commerce habits of the French. The French spent 25.9 billion euros, which corresponds to an increase of 5.3%. This overall growth is very contrasted depending on the purchases made. All products and services combined recorded 408 million online transactions. There has been a sharp drop in purchases of services partly offset by a substantial increase in product sales. The sale of consumer products increased by 45.7%. The average basket has increased and exceeds 60 euros, i.e., a 6.8% increase compared to the 2nd quarter of 2019. Product sales represent 57% of overall revenue (compared to 44% on average in 2019). Online purchases from store banners increased by 83%. 85% of Internet users have opted for home delivery. Mobile was less used than computers to place online orders, with just 23% of mobile shoppers during the first lockdown, compared to 69% of desktop shoppers and the remaining 8% on tablets. Unsurprisingly, the e-commerce giants (Amazon, LeBoncoin, or Cdiscount) are still at the top of the ranking of the most successful e-commerce sites. Their secret to maximizing their sales; A wide choice of products, fast delivery, and attractive prices. According to Comscore, 50% of Internet searches will soon use voice search. Consumers are increasingly using it for their purchases. That poses many new challenges and opportunities for e-commerce players who will have to structure the content of their merchant site accordingly. There is also a trend to simplify merchant sites with less
28
A. Samih et al.
sophisticated interfaces. The goal is to put the product back at the customer experience center by enhancing it. The user-tested numerous sites during the lockdown; he knows what to expect in the shopping experience and admits to being disappointed. E-commerce sites must enhance the user experience by providing all the technology required to make finding and purchasing things easier, faster, and more efficient; the shopping experience on e-commerce platforms must be improved. Customers demand trustworthy information at every trip level, from product search to checkout. They want to go through the process stages quickly and without re-enter data. They also want a variety of payment options. The lowest two categories that shopped households online in Canada before the pandemic were singles and single mothers are living with other adults (such as parents and housemates) [39]. These groups accounted for 12% and 3% of all users, respectively. Some of these groups have users who have never shopped online or used e-commerce for different needs. In Canada, millennials and baby boomers had the highest percentage of online consumer sales in 2019 (since 2020). Today, the preference is for homeschooling and ordering what is needed to entertain kids. A 2019 report by Canada Post indicated that 62% of Canadians shopped for clothing apparel, whereas 41% shopped for computers and electronics using e-commerce. After the pandemic hit, the Cision (2020) report showed that all e-commerce sales increased, except for clothing (the lowest increase of 21%). Meanwhile, the sales of electronics increased by 160% [40]. Consumers aged 73 and over and customers aged 18–23 account for the smallest percentages of online shoppers in Canada (5% in each category) in 2019, as shown in Fig. 2. These clients represent two distinct generations, making it harder to target them. According to Forbes (2020), technology is critical for improving the Gen Z buying experience and offering them quick and high-quality services [41]. By 2026, it is expected that 21% of the population in Canada will be 73 years old or older. Grey-sheep customers are ordinary, and as described in [42], they favor products that improve their quality of life. According to [42], this category of clients prefers to see things in person before purchasing them [42]. For example, when Canada’s pandemic lockdown began in mid-March 2020, several food businesses set aside specific hours for older consumers. However, according to recent research by Statistics Canada, a considerable percentage of these customers turned to internet shopping, with 45 percent of respondents aged 75 and up saying they did so [43]. The difficulty is how to target these customers and, given that many have turned to e-commerce for the first time, how to maintain them as customers once life returns to normal. The following section showed how to apply adaptable recommendation algorithms to solve this problem.
5 Proposed Solution Based on Adaptive Recommendation Algorithms Due to the corona health crisis, recommendation systems must be based on new methods. • Multicriteria methods, mainly since several users use these systems for the first time.
The Impact of Covid 19 on Recommendation Platforms
29
• Multidimensional methods or psychological notions such as emotions, opinions, or contextual “age, place, time…”. For example, news websites are often constantly updated; some articles may be replaced by a “breaking news” article on the same topic several times during the same day, which may require constant updates of the recommendation models. Another typical challenge in the information field is that a user’s interest can change dynamically, depending on different contextual factors, such as the time of day, the characteristics of their device (e.g., the mobile phone), his age. Or the user’s current location (office). Before we talk about the context in recommender systems, we should know the context in general. The definition of context in the Webster’s Dictionary is as follows: “the parts of speech which surround a word or passage and can illuminate its meaning; condition or circumstances that have an effect; the interrelated conditions in which something exists or occurs: environment, the definition of the historical context of the war. " Recommendation systems traditionally operate on a "User Item" matrix. An attribute vector can describe each user, called a user profile. The matrix becomes "User Item Context" when the context is added. The integration of context in recommendation systems adds a dimension of complexity to the model. Recommendation data, as the assessments may be valid in only one particular context. The context can be described by a vector of context attributes, such as time, location, or currently available network bandwidth in a mobile scenario. The context attributes used to depend on a large part of the application area.
Fig. 3. Drifts in user behavior (using [45, 50]).
However, methods using so-called psychological knowledge are also of great importance because the crisis has completely changed the psychology of clients. Note that
30
A. Samih et al.
psychological effects can be described by psychological states such as emotions, persuasion, attention, presence, etc… These effects are then used to predict the person’s psychological state using a system at a specific time, influencing the results obtained in recommender systems. All recommendation systems strive to create a relationship of trust with the user. However, many psychological interactions occur when users simultaneously access a website where a referral system resides. These interactions and the importance of the various psychological components are not easy to understand. Most of the actions of users who go to a website and make decisions come from sections of the brain that are in the subconscious; So people do not know and cannot say why they took the actions they took. Therefore, developers of recommender systems need to study psychological aspects to understand these interactions to create a system as effectively as possible. Recommendation algorithms must address the "why" problem to withstand the intense competition between recommendation platforms; these techniques provide recommendations to users and explanations, allowing the system user to understand why such items are recommended. In this way, the system help improves the effectiveness, efficiency, persuasion, and satisfaction of users. Thus, the explainable recommendation aims to design interpretable models that function humanly, and such models generally lead to explaining the recommendation results. Most recommendations researchers today fall into this category, which aims to understand how recommending processes work, and it typically refers to a model-based explainable recommendation. Recommendation systems are among the most successful and widespread uses of machine learning technologies. In fact, The COVID-19 epidemic causes a significant and abrupt shift in consumer behavior. Furthermore, the presence of a large number of new and atypical consumers posed a new recommendation challenge. Traditional machine learning algorithms cannot automatically detect and handle such changes in user preferences and profiles. This means a reduction in the accuracy of their predictions, which indicates that the models are inaccurate, necessitating the creation of new models; this usually occurs after some time has passed. Online or incremental learners, such as trees and adaptive sets, are well-suited to learning in such dynamic situations [44, 45]. These incremental learning algorithms update their models as new data arrives. They can ’forget’ old notions by utilizing replacement, which involves deleting unnecessary model paragraphs and replacing them with subsections trained on recent data. Based on [45]; This procedure is enabled by implementing drift detection into their designs; as a result, these algorithms can modify their models dynamically and smoothly to changes in user preferences. Although direct concept drift detection is not required for incremental algorithms to adjust to drifting concepts, as they frequently do so naturally by updating and forgetting, it does provide several benefits. If idea drift occurs suddenly, the model can quickly recognize and adjust to it. Concept drift identification also provides insights into the mechanics of the generation process, making future recurrent or seasonal variations easier to model (see Fig. 3). On the COVID-19 pandemic, it is too early to tell if a fourth or fifth wave will occur. Nor is it possible to predict consumer behavior in the unfortunate event that these waves
The Impact of Covid 19 on Recommendation Platforms
31
arise. The use of e-learning algorithms that integrate drift detection algorithms appears to be a research direction. So the objective will be the creation of a recommendation system that can use contextual, psychological, and explainable methods "why the system recommended these items for you" to achieve maximum user confidence and satisfaction. This fusion between online learning algorithms based on drift detection and the above methods help to ensure that e-business can adapt quickly and effectively to changes in their customers and their purchasing habits [45] while facilitating interactions with cold starts and grey sheep.
6 Conclusion With the advent of the Coronavirus, today we are witnessing a dramatic change in users’ lifestyles. As well as new consumers started to use these systems for the first time, a large percentage of them have preferences outside the regular (grey sheep user). This exponential change translates into difficulty organizing and analyzing this basic information but opens new avenues on the paths of knowledge. The question is no longer to have the information but to find the relevant information at the right time regardless of the type of user. In this paper, we proposed to integrate a fusion of new methods in the recommendation process to solve the challenges posed by the health crisis; we offered to use multicriteria, multidimensional, psychological, explainable techniques (explain the recommendations), and online machine learning algorithms that assure the adaptability according to user situation and type.
References 1. Samih, A., Ghadi, A., Fennan, A.: ExMrec2vec: explainable movie recommender system based on Word2vec. Int. J. Adv. Comput. Sci. Appl. (IJACSA) 12(8), (2021). https://doi.org/ 10.14569/IJACSA.2021.0120876 2. Samih, A., Ghadi, A., Fennan, A.: Deep graph embeddings in recommender systems: a survey. J. Theor. Appl. Inf. Technol. 99(15) (2021). https://doi.org/10.5281/zenodo.5353504 3. Samih, A., Ghadi, A., Fennan, A.: Translational-randomwalk embeddings-based recommender systems: a pragmatic survey. In: Kacprzyk J., Balas V.E., Ezziyyani M. (eds.) Advanced Intelligent Systems for Sustainable Development (AI2SD’2020). AI2SD 2020. Advances in Intelligent Systems and Computing, vol. 1418. Springer, Cham (2022). https:// doi.org/10.1007/978-3-030-90639-9_77 4. Samih, A., Adadi, A., Berrada, M.: Towards a knowledge based explainable recommender systems. In: Proceedings of the 4th International Conference on Big Data and Internet of Things, Rabat Morocco, pp.1–5, October 2019. https://doi.org/10.1145/3372938.3372959 5. Bhatti, A., Akram, H., Basit, H.M., Khan, A.U., Raza, S.M., Bilal, M.: E-commerce trends during COVID-19 pandemic. Int. J. Future Gener. Commun. Networking 13(2), 5 (2020) 6. Borojo, D.G., Yushi, J., Miao, M.: The effects of COVID-19 on trade, production, environmental quality and its implications for green economy. J. Econ. Stud. (2021). https://doi.org/ 10.1108/JES-06-2021-0307 7. Mirbakhsh, N., Ling, C.X.: Improving top-N recommendation for cold-start users via crossdomain information. ACM Trans. Knowl. Disc. Data (TKDD) 9(4), 33 (2015)
32
A. Samih et al.
8. Alabdulrahman, R., Viktor, H.: Catering for unique tastes: targeting grey-sheep users recommender systems through one-class machine learning. Expert Syst. Appl. 166, 114061 (2021) 9. Hawashin, B., Alzubi, S., Mughaid, A., Fotouhi, F., Abusukhon, A.: An efficient cold start solution for recommender systems based on machine learning and user interests. In: 2020 Seventh International Conference on Software Defined Systems (SDS), pp 220–225. IEEE (2020) 10. Venkatesan, R., Sabari, A.: Issues in various recommender system in e-commerce – a survey. J. Crit. Rev. 7(7), (2020). ISSN-2394-5125. https://doi.org/10.31838/jcr.07.07.109 11. Huang, Y., Liu, H., Li, W., Wang, Z., Hu, X., Wang, W.: Lifestyles in Amazon: evidence from online reviews enhanced recommender system. Int. J. Market Res. 62(6), 689–706 (2020). https://doi.org/10.1177/1470785319844146 12. Gomez-Uribe, C.A., Hunt, N.: The Netflix recommender system: algorithms, business value, and innovation. TMIS 6(4), 13 (2016) 13. Jannach, D., Manzoor, A., Cai, W., Chen, L.: 2021. A survey on conversational recommender systems. ACM Comput. Surv. (CSUR) 54(5), 1–36 (2021) 14. Anelli, V.W., Deldjoo, Y., Di Noia, T., Ferrara, A., Narducci, F.: FedeRank: user controlled feedback with federated recommender systems (2021). arXiv:2012.11328. [csIR] 15. Adadi, A., Berrada, M.: Peeking inside the black-box: a survey on explainable artificial intelligence (XAI). IEEE Access 6, 52138–52160 (2018). https://doi.org/10.1109/ACCESS. 2018.2870052 16. Portugal, I., Alencar, P., Cowan, D.: The use of machine learning algorithms in recommender systems : a systematic review. IEEE J. 97, 2–11 (2018) 17. Deldjoo, Y., Schedl, M., Knees, P.: Content-driven Music Recommendation: Evolution, State of the Art, and Challenges. arXiv preprint arXiv:2107.11803 (2021) 18. Mokarrama, M.J., Khatun, S., Arefin, M.S.: A content-based recommender system for choosing universities. Turkish J. Electr. Eng. Comput. Sci. 28(4), 2128–2142 (2020) 19. Sharma, A., Jiang, J., Bommannavar, P., Larson, B., Lin, J.J.: GraphJet: real-time content recommendations at Twitter. PVLDB 9 13(2016), 1281´s1292 (2016) 20. Zhu, Z., He, Y., Zhao, X., Zhang, Y., Wang, J., Caverlee, J.: Popularity-opportunity bias in collaborative filtering. In: Proceedings of the 14th ACM International Conference on Web Search and Data Mining, pp. 85–93 (2021) 21. Jain, G., Mahara, T., Tripathi, K.N.: A survey of similarity measures for collaborative filteringbased recommender system. In: Pant, M., Sharma, T.K., Verma, O.P., Singla, R., Sikander, A. (eds.) Soft Computing: Theories and Applications. AISC, vol. 1053, pp. 343–352. Springer, Singapore (2020). https://doi.org/10.1007/978-981-15-0751-9_32 22. Kulkarni, P.V., Rai, S., Kale, R.: Recommender system in elearning: a survey. In: Proceeding of International Conference on Computational Science and Applications, pp. 119–126 (2020) 23. Kumar, P., Thakur, R.: Recommendation system techniques and related issues: a survey. Int. J. Inf. Technol. 10, 495–501 (2018) 24. Benouaret, I., Lenne, D.: A composite recommendation system for planning tourist visits. In: IEEE/WIC/ACM International Conference on Web Intelligence (2016) 25. Zhao, C., Li, C., Xiao, R., Deng, H., Sun, A.: Catn: cross-domain recommendation for coldstart users via aspect transfer network, arXiv preprint arXiv:2005.10549 (2020) 26. Argaman, O.: Why customer re-engagement has become more valuable during COVID19 (2020). https://www.destinationcrm.com/Articles/WebExclusives/Viewpoints/Why-Cus tomer-Reengagement-Has-Become-More-Valuable-DuringCOVID-19-141234.aspx 27. Lu, J.: Recommender system application developments: a survey. Decis. Support Syst. 74(12), 32 (2015)
The Impact of Covid 19 on Recommendation Platforms
33
28. Lamche, B., Trottman, U., Worndl, W.: Active learning strategies for exploratory mobile recommender systems. In Proceedings of the fourth Workshop on Context-Awareness in Retrieval and Recommendation (2014) 29. Jindal, H., et al.: PowKMeans: a hybrid approach for gray sheep users detection and their recommendations. Int. J. Inf. Technol. Web Eng. 13(2), 56–69 (2018) 30. Zheng, Y., Agnani, M., Singh, M.: Identifying grey sheep users by the distribution of user similarities in collaborative filtering. In: Proceedings of The 6th ACM Conference on Research in Information Technology. ACM (2017) 31. Abdulrahman, R., Viktor, H.: Catering for unique tastes: targeting grey-sheep users recommender systems through one-class machine learning. Technical Report IDEAL1–2020, University of Ottawa (2020) 32. Elrhim, M.A., Elsayed, A.: The effect of COVID-19 spread on the e-commerce market: the case of the 5 largest e-commerce companies in the World. Available at SSRN 3621166 (2020) 33. Fevad. L’impact de la crise Covid-19 sur les achats e-commerce en B2B (2021). https://www. fevad.com/limpact-de-la-crise-covid-19-sur-les-achats-e-commerce-en-b2b/ 34. Kim, R.Y.: The impact of COVID-19 on consumers: preparing for digital sales. IEEE Eng. Manage. Rev. 1, 1 (2020) 35. Bouaziz, D.: Quel est l’impact du coronavirus dans l’e-commerce (2020). https://www. ecommercemag.fr/Thematique/data-room-1223/Breves/Quel-est-impact-coronavirus-e-com merce-347825.htm 36. Ungerer, C., Portugal, A., Molinuevo, M., Rovo, N.: Recommendations to leverage ecommerce during the covid-19 crisis. In: World Bank Group (2020) 37. BBC News, Coronavirus: quatre façons dont Covid a changé notre façon de vivre - et de faire du shopping (2020). https://www.bbc.com/afrique/monde-55400771 38. Fevad 2020, Les chiffres clés et tendances d’achat du e-commerce (2020). https://www.camoncommerce.com/aide/guide-actus/blog/nos-guides-e-commerce/actualites-e-commerce/ les-chiffres-cles-et-tendances-dachat-du-e-commerce/ 39. Statista. Distribution of online shoppers in Canada as of April 2019, by household type. Cited 2020 (2019). https://www.statista.com/statistics/1044443/canadaonline-shoppers-by-househ old-type/ 40. Absolunet. COVID-19 driving consumers online: Canadian ecommerce sales double in 2 weeks. In CISION (2020) 41. Goldstein, J.: What does Gen Z want from retail? Forbes (2020). https://www.forbes.com/ sites/forbesbusinessdevelopmentcouncil/2020/02/25/what-does-gen-z-want-fromretail/#578 ce084661b 42. Insider, R.: How does aging change Canada’s consumer behavior? Retail Insider Press (2020). https://www.retail-insider.com/articles/2020/2/howdoes-aging-change-canadas-con sumer-behavior 43. Post, C.: The 2020 Canadian E-commerce Report. Canada Post Press, Ottawa (2020) 44. Bifet, A., Gavalda, R., Holmes, G., Pfahringer, B.: Machine Learning for Data Streams with Practical Examples in MOA. MIT Press, Cambridge (2018) 45. Abdulrahman, R., Viktor, H.L.: Personalised recommendation systems and the impact of COVID-19: perspectives, opportunities and challenges, pp. 295–301 (2020) 46. Pricecomparator, L’impact DU coronavirus en e-commerce (2020). https://pricecomparator. pro/impact-coronavirus-e-commerce/ 47. Sommer, P.L: L’impact du Coronavirus sur l’e-commerce en France (2020). https://contentsq uare.com/fr-fr/blog/ecommerce-coronavirus-barometre-semaine-12/ 48. Sielis G.A., Tzanavari, A., Papadopoulos G.A.: Recommender systems review: types, techniques and applications. In: Encyclopedia of Information Science and Technology, 3rd edn, Chapter 714, pp. 329–339 (2014)
34
A. Samih et al.
49. Seyednezhad, M., Cozart, K., Bowllan, J., Smith, A.: A review on recommendation systems: context-aware to social-based, IEEE, pp. 9–20 (2018) 50. Gama, J., Zliobaite, I., Bifet, A., Pechenizkiy, M., Bouchachia, A.: A survey on concept drift adaptation. ACM Comput. Surv. 46(4), 1–37 (2014)
Classifiers-Based Personality Disorders Detection Fatemeh Sajadi Ansari1,2 , Djamal Benslimane2(B) , Aymen Khelifi1 , and Mahmoud Barhamgi2 1
Kaisens Data, 9/11 Allee de L’Arche, 92400 Courbevoie, France 2 Claude Bernard Lyon 1 University, Lyon, France [email protected]
Abstract. Internet technologies including social networks allow users to easily communicate with each other. This provides us an interesting resources space to early detectection of abnormal behavior such as mental disorders. Important mental factors were initially proposed in some psychological solutions. Recently, machine learning-based apporaches are proposed and tend to exploit the large data that social networks can provide to detect abnormal behavior in its early stage. In this paper, we propose a set of classifiers-based troubles detection in the context of social media. Message toxicity detectors, gender classifiers, age estimators, and personality predictors are some examples of the various classification modules employed in our approach. The combined results of these classifiers can be used to uncover suspicious activity in user profiles. Our difefrent classifiers are trained by using different datasets in the context of twitter and instagram platforms. Keywords: Abnormal Behavior Media · Big Fives · Classifiers
1
· Personality Features · Social
Introduction
Advances in mobile and Internet technologies such as smart phones, and social media have truly transformed our lives and reshaped the way we interact with each other and socialize. The arrival of fast 5G networks and the Internet of Multimedia Things (IoMT) is even expected to take this transformation to an advanced level where people can communicate and exchange multimedia content with each other through a multitude of smart interfaces embedded in their surrounding physical objects (e.g., cars, fridges, etc.). However, while these technologies promise to ease our lives, they also come with a wide range of undesirable side effects such as Nomophobia, mental disorders [1], network addictions and cyberbullying [2]. Psychological solutions were initially proposed to mainly identify important mental factors. Recently, new machine-learning based apporaches were proposed to exploit the large data available on social media to detect mental disorders at its early stage. c The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 35–45, 2023. https://doi.org/10.1007/978-3-031-35248-5_4
36
F. Sajadi Ansari et al.
In this research, we propose a method for identifying psychological distress that relies on machine learning. Different unit classifiers are proposed and integrated together. The substance of messages may not be adequate to identify situations with personality problems, in our opinion. It should be supported by studying the probable personality difficulties scenario from other angles. Toxic message exchange detection, gender prediction, age estimation, and big five model-based personality estimate are the four unit classifiers brought together here. The rest of this paper has the following structure. Related work is discussed in Sect. 2. In Sect. 3, we provide a high-level description of the methodology and its individual parts. Section 4 gives an overall picture of a prototype, presents experiments and discusses some results. The final section of the paper summarizes the main points.
2
Related Work
A considerable work has been devoted to mental disorders and cyberbullying detection in the context of social media. In [1], authors argue that mining behaviors from social media is one way to identify some social network mental disorders such as net compulsion and cyber-relationshp addiction. They propose a machine-learning based approach to identify potential cases of mental disorders. Mental Disorder Detection is considered as a semi-supervised classification problem and three types of disorders are detected with a binary tranductive SVM: Cyber-Relationship Addiction, Net Compulsion, and Information Overload. The proposed approach is mainly based on extracting and exploiting different features of users including social structure, online and offline interaction ratio, and social diversity. Tensor techniques are also exploited to better infer teh user behavior from logs of differents social media platforms. In [3], authors consider that mental disorders occur often in combination and are then related to each other. Several online mental healthrelated communities are studied to extract topics and psycho-linguistic features with an interest on depression from a large number of posts (620K) initiated by a huge number of users (80K). These features are used as an input of a join modelling framework to classify mental health-related communities. The proposed framework’s efficacy was evaluated against single-task logistic regression and a multi-task learning model. In [4], authors explore social media usage and statistical classifier technique to detect and diagnose at eraly stage depressive disorder on Twitter platform. They rely on crowdsourcing technique to compile a set of Twitter users who reported being diagnosed with clinical depression. Posts are analyzed for a variety of behavioural characteristics in the year leading up to depression. These characteristics are associated with indicators of social participation (post frequency, percentage of replies, etc.), emotional states (positive and negative affect, activation and dominance), and Depression Language (Depression lexicon and Antidepressant usage) Authors highlighted victim conduct as a cyberbullying signal in [5,6]. Victims’ early responses, such as postings and statuses on social
Classifiers-Based Personality Disorders Detection
37
networks, may serve as a warning indicator of their emotional and psychological state when cyberbullied. In a similar vein, the authors of [7] showed how effective a Decision Tree classifier could be in detecting cyber bullies by taking into account factors other than the content of messages, such as the users’ demographics and social dynamics. The authors of [8] investigated the use of deep learning models to the problem of identifying instances of cyberbullying. When both types of models were applied to the same dataset, the compassion study found that the deep learning-based models performed significantly better than the traditional machine learning models. An expert solution was presented by the authors in [9] to identify bully individuals on social media. The suggested method integrates the knowledge and experience of cyberbullying professionals (humans) with the findings of supervised machine learning models to reach a greater accuracy in identifying bully users. The goal of this method is to prevent cyberbullying. The experiments demonstrated that this expert system is superior to supervised machine learning models.
3
Our Approach
3.1
Approach Overview
Figure 1 provides a high-level overview of our process. Our method applies data mining and machine learning algorithms to assess a user’s profile and evaluate the extent to which it exhibits “unconventional” characteristics. Our system aims to detect any minuscule signal that may indicate the presence of a cyberbullying danger. Our method is multimodal, consisting of four individual modules that work together to get a social network profile dataset ready for profile clustering analysis. The objective of the first subprocess is to prepare the indicators associated with the profile’s emitted contents. We have trained algorithms for categorizing harmful content because we intend to detect anomalous behavior to determine whether a profile constitutes a concern. This technique also calculates psychological content features to identify linguistic patterns that signify personality disorder signals, such as paranoid, borderline, and narcissistic. To achieve this, we have completed the process of generating the necessary datasets to train and optimize our multiple classifiers. As was demonstrated in the module titled “Data Collection and Processing” found in Fig. 1, not only do we collect actual data from social media platforms such as Twitter and Instagram, but we also end up making use of available data and corpuses that the scientific community1,2 utilize. All of the data obtained, including Tweets and Instagram posts, is scrubbed and anonymized so that it does not include any personally identifying information. This is done to guarantee that all applicable data protection regulations are met (e.g., GDPR).
1 2
https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/data. https://www.chatcoder.com.
38
F. Sajadi Ansari et al.
Fig. 1. An overview of the proposed approach for profile clustering
The second sub-process calculates frequency indicators such as the average publication per day or the publication rates by time slots. These frequencies may be representative of addictive practices, stress or sleep disorders. The third sub-process works out profile’s network indicators such as number of subscribers and number of subscriptions. A demographic indicator is calculated on the basis of a gender classification. According to some psychological studies, gender correlates with certain user behavior on social networks. Finally, based on the textual content of the profile during the previous six months, the fourth sub-process evaluates personality traits using the “Big Five” model. The following are the three personality qualities that we considered: omn/ext/neu. These 4 sub-processes allowed us to calculate 12 indicators from which, we modeled a profile and the distance between the profiles. Each of these categories of indicators can intervene in different aspects of modeling. Table 1 illustrate a summary of indicators calculatd for each profile collected and the axis of the modeling. analysis of correlation between these different indicators led us to focus on 4 axes for clustering. Our ensuing sections go into greater depth regarding these various procedures.
Classifiers-Based Personality Disorders Detection
39
Table 1. indicator interpretations Indicator
Interpretation
Number Of Subscribers
Allows Us To Verify The Legitimacy Of The Profile.
Number Of Subscriptions
Detects Bots.
Biography - Description
Intervenes In Toxicity And Personality.
Average Daily Publication
Helps Detect Addictive Behavior.
Average Publication By Time Slot Detects Addictions And Abnormal Behaviors. Average Reactions Per Publication Helps To Detect Spam. Average Toxic Comments Per Post Detects The Toxicity Of The Profile. Age
Determines Trends By Age.
Gender
Determines Trends By Gender.
Shared Texts
Intervenes In The Calculation Of Toxicity And Personality.
Overall Toxicity Rate
Determines Toxic Trends.
Personality Traits
Helps Identify Mental Disorders Specially Neuroticism.
3.2
Unit Classification Modules
Multiple distinct classifiers serve as the backbone of our methodology. In addition to message content, other demographic variables, such as gender, can assist in identifying personality disorders. The three primary system classifiers given are the content classifier, the gender estimation classifier, and the personality classifier. Textual Categorization: The cyber harassment semantic domain enables the creation of a multimedia predictor that may identify harassing, sexually predatory, offensive, racist, or sexist information. A set of linguistic or spatial indications is computed and utilized with specific dictionaries and lexicons to identify potentially harmful input in interpersonal conversations (Sect. 4.2). Relying on annotated datasets that align with an ontology of online bullying, this categorization provides a more specific toxicity level, which corresponds to the category of harassment. Due to this more comprehensive description of toxicity, we will more precisely assess the danger posed by newly found harmful substances. Male-Female Recognition: Its purpose is to recognize the gender of social network profiles through the content emitted. Furthermore, it seems that gender identification is simple for humans but challenging for machines. Face features, speech characteristics, electroencephalogram readings, etc., and sociodemographic characteristics (handwriting, blog, ...) can help researchers determine a person’s gender. Our method uses information from several social networks to achieve automatic gender classification. Personality Estimation: we use this classifier to estimate the essential traits of a social network user profile. One of the most widely-applied models, the “big five” [10], is employed here. Human differences in thought, emotion, and
40
F. Sajadi Ansari et al.
behavior models are referred to as “personality.” A person’s general traits, such as mood, attitude, and personality, are combined to form their personality. Personality traits and qualities have a big influence on our lives; they determine our lifestyle, health, and other aspirations. Consequently, the personality of each individual could influence his behaviors including his activities on social networks. We focused on 3 personality traits that According to Power RA. [11] are defined as follows: Neuroticism: symptoms include melancholy, impatience, and emotional instability. People who are highly sensitive tend to experience mood swings, anxiety, impatience, and sadness. Those who are extraverted are those who are easily energized, who enjoy interacting with others, who are outgoing and social, who are confident in themselves, and who express their feelings strongly (or extraversion). Those who score high on the extraversion scale are the types of people who enjoy being in the company of others and who perform well in social situations. Creativity and introspection are hallmarks of openness. Those who score higher on this dimension are more likely to take risks and experiment with new ideas. People that are low on this feature tend to be more conventional and may have trouble with abstract ideas.
4
Prototype, Experiments and Results
4.1
Datasets Description
In this section we will briefly describe the different datasets used for clustering profiles, toxicity, gender and personality classifying. Clustering Dataset: Based on the list of Instagram profiles in the Askfm dataset3 , we collected posts and comments from over 100 Instagram profiles. we were only interested in textual publications, not Reels or Stories, which are in the field of computer vision. Then using a keyword list we built a Twitter profile list from which we collected all the associated posts and comments of 150 profiles. To construct a dataset containing the profiles of the two social networks of Twitter and Instagram, we took a set of common information between the two social networks. We applied a mapping between the two types of attributes. Table 1 summarizes the final list of information common to Instagram and Twitter. Toxicity Classification Dataset: We concatenated several existing datasets to build a rich dataset with our defined toxicity classes. Then to balance our dataset, we proceeded to the data augmentation but also to the collection of data on social networks. The list of existing datasets used to train the toxicity classification is illustrated in Table 2.
3
https://sites.google.com/site/cucybersafety/home/cyberbullying-detectionproject/dataset.
Classifiers-Based Personality Disorders Detection
41
Table 2. Existing datasets for toxicity classification Class of toxicity
Link to data sources
Insulting remarks
https://www.kaggle.com/c/detecting-insults-in-socialcommentary/data?select=train.csv https://www.kaggle. com/mrmorj/hate-speech-and-offensive-language-dataset
Threatening remarks https://www.kaggle.com/c/jigsaw-toxic-commentclassification-challenge Defamatory remarks https://competitions.codalab.org/competitions/26654# learn the details\discretionary-dataset Obsene remarks
https://github.com/abhilashabhagat/Obscene-detection/
Racial remarks
https://github.com/HannahKirk/Hatemoji https:// zenodo.org/record/3520152#.Yjsulef MJPZ
4.2
Experimental Evaluation
As shown in Table 3, we used five different classifiers. Some of these classifiers are more complex than others (such as identifying toxic remarks or determining a user’s age). On the other hand, toxic tweet detection and gender prediction are binary classifications. Table 4 presents the findings of a personality analysis performed on the basis of the information provided in the supplied text. Our model that predict the gender of a person from their first name by using BERT gender classifier, has an accuracy rate of 84% overall. Our Camembert multiclasses classifier-based toxicity detection achieves an accuracy of 83% and an F-score of 85% in terms of its performance. The evolution of the accuracy of toxicity multiclasifier and gender prediction are illustrated in Fig. 5. The vectorization of our dataset was accomplished through the utilization of the TF-IDF (Term Frequency-Inverse Document Frequency) word-importance measure as well as the word-embeddingbased bidirectional BERT model. 4.3
Clustering and Feature Selection
Implementation of the unitary classifiers allowed the calculation of the indicators we identified as crucial dimensions for social networks’ user profile modeling. We have presented a mapping of the latter to determine which indications are associated with clusters forming. To better visualize the structure of the two clusters, we introduced the indicators one by one and created correlation pairs between them. We could see which signs the two clusters emerged from using this mapping. This step will intervene in defining key dimensions for social network profile modeling. On the one hand, the two clusters are formed at the intersection of publication habits according to time slots and the axis of toxicity, as shown in the Fig. 2. They are, on the other hand, made up of the two personality qualities of extravagance and neuroticism, as well as publishing habits based on time slots.
42
F. Sajadi Ansari et al.
Fig. 2. feature correlations
We were able to establish the factors that had the strongest correlation with each other after multiple tests. The following factors are identified as most significant: (a) Average number of posts each day, (b) Average number of posts per time slot, (c) Personality profile, (d) Toxicity rate on a global scale.
Classifiers-Based Personality Disorders Detection
43
Using the K-means technique, we were able to confirm the existence of two (2) separate clusters, which were, nevertheless, difficult to visualize due to the large number of criteria considered. 4.4
Clusters Interpretation
Visualization of Fig. 3 enables us to make the first observation. The y-axis depicts the rate of publication throughout the night, the X axis depicts the overall toxicity rate, and the size of the bubbles characterizes the profiles’ neuroticism. We can see two clusters in this graph: the first is distinguished by a large number of users profiles belonging to it and a meager of publication at night and a very low rate of anxiousness. The second, on the other hand, is defined by the small number of people who belong to it, and is characterized by a high rate of publication at night and a strong neurotic attitude. We can have a better understanding of the phenomenon by means of a second Visualization. The association between anxiousness, toxicity, and publishing behaviors. Two groups can be seen forming along the axis of toxicity, as illustrated in the figure Fig. 4. We notice a tiny number of people whose publishing habits are considerably more focused on the night and whose toxicity is much more than their level of anxiousness, which appears to be nonexistent or very near to nil. According to the results from these visualizations of profile’s activities collected, we were able to establish some assumptions and arrive to the following conclusions: There is a substantial link between late-night posting behaviors and neuroticism. The majority of the population studied has a weak tendency for nocturnal publication and a weak neuroticism tendency, while a small portion of the population has a mixed nightly posting pattern and high level neuroticism. The second group identified has a strong tendency for toxicity in exchange for a very weak neurotic tendency, leading us to believe that the toxicity detected is hypothetically intentional. On this basis we concluded that profiles emerging towards the right-wing cluster may represent a greater risk of being a harasser because they approach the definition of harassment that is intentionally hurting a person who can hardly defend himself.
Fig. 3. Relationship between night publication habits and toxicity
44
F. Sajadi Ansari et al.
Fig. 4. Relationship between publication habits, toxicity and profile neuroticism
Fig. 5. Toxicity and Gender detection Accuracy Table 3. Main implemented classifiers Natural Language intention of the model
Class Labels
ML models
English
Toxic comments detection
Toxic, severe toxic, Bert insult, obscene, identity hate
English
Gender prediction from text
Male, female
English
Gender prediction from name Male, female
LSTM, CNN
English
Personality analysis
Big 5 labels
SVM, Bert
french
toxic tweets classification
Toxic, non toxic
CamemBERT
SVM, Bert
Table 4. Personality analysis from text Value
Big5 Op Co
Ag
Ex
Ne
Boolean value True True True False False probability
0.65 0.56 0.66 0.56
0.29
Classifiers-Based Personality Disorders Detection
5
45
Conclusion
In this paper, we proposed an approach to automatically detect potentiel users with abnormal behavior. Our approach relies on machine learning techniques and is based on combining diverse classifiers. It can works for different social networks but separately. We have just used textual posts thus far in our exploits. Next, we plan to investigate how social network characteristics influence our models so that we don’t overlook any potential instances of anomalous behavior. The user interaction graph could be mined for such details. we also intend to extend our solution to include multimedia posts by using advanced techniques of computer vision and embedding techniques. We finally project to study the user behavior by integrating the posts and the metadata that can be extracted not only from one social media but from different social media.
References 1. Shuai, H., et al.: A comprehensive study on social network mental disorders detection via online social media mining. IEEE Trans. Knowl. Data Eng. 30(7), 1212– 1225 (2018) 2. Mediacorp, “3 in 4 youngsters say they have been bullied online (2018) 3. Saha, B., Nguyen, T., Phung, D.Q., Venkatesh, S.: A framework for classifying online mental health-related communities with an interest in depression. IEEE J. Biomed. Health Inf. 20(4), 1008–1015 (2016) 4. De Choudhury, M., Gamon, M., Counts, S., Horvitz, E.: Predicting depression via social media. In: Proceedings of the Seventh International Conference on Weblogs and Social Media, ICWSM 2013, Cambridge, Massachusetts, USA, July 8-11, 2013, Kiciman, E., Ellison, N.B., Hogan, B., Resnick, P., Soboroff, I., (eds.) The AAAI Press (2013) 5. Choudhury, M.D., Counts, S., Horvitz, E.: Social media as a measurement tool of depression in populations. In: the Web Science Conference, pp. 47–56 (2013) 6. Dadvar, M., Ordelman, R., de Jong, F., Trieschnigg, D.: Towards user modelling in the combat against cyberbullying. In: 17th International Conference on Applications of Natural Language to Information Systems, pp. 277–283 (2012) 7. Squicciarini, A.C., Rajtmajer, S.M., Liu, Y., Griffin, C.: Identification and characterization of cyberbullying dynamics in an online social network. In: The IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, ASONAM, vol. 2015, pp. 280–285 (2015) 8. Dadvar, M., Eckert, K.: Cyberbullying detection in social networks using deep learning based models. In: Song, M., Song, I.-Y., Kotsis, G., Tjoa, A.M., Khalil, I. (eds.) DaWaK 2020. LNCS, vol. 12393, pp. 245–255. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-59065-9 20 9. Dadvar, M., Trieschnigg, R., de Jong, F.: Experts and machines against bullies: a hybrid approach to detect cyberbullies. In: 27th Canadian Conference on Artificial Intelligence, vol. 5, pp. 275–281 (2014) 10. John, O.P., Srivastava, S.: The big five trait taxonomy: History, measurement, and theoretical perspectives. In: Handbook of personality: Theory and research, vol. 2, pp. 102–138 (1999) 11. Power, R., Pluess, M.: Heritability estimates of the big five personality traits based on common genetic variants. Transl. Psychiatry 5, e604 (2015)
How Health Information Technology Improved Patient Care and Treatment During the COVID-19 Pandemic: A Comparison Between International Case Studies and the Moroccan Context Ahmed Kadiri(B) , Hamid Azzouzi, and Noufel Sefiani Equipe de Recherche, Ingénierie, InnovationEtManagement Des Systèmes Industriels, FST Tanger, Tangier, Morocco [email protected]
Abstract. In the healthcare industry, health information technology (HIT) refers to the notion of cloud-based services, such as the internet, linked networks, and so on. It primarily makes use of electronic medical records, patient information, and data to deliver more efficient and sophisticated treatments and services. The aim of the research is to give roles and potential applications and also evaluate the influence of the HIT concept on care delivery during the existing COVID-19 epidemic. The scope of HIT in the health care industry and how countries such as Morocco can benefit from an international case study in a push to overhaul and modernize their hospital information systems. The writers perform their study with the use of reliable data sources such as IEEE, Springer, Elsevier, Taylor & Francis, Sprouts, and Google Books, as well as surveys of subject area experts. The authors used a methodical approach to study and diagnose the development of the hospital process digitization project in the Moroccan environment, taking into account past publications and studies on the subject, as well as an effective survey directed towards healthcare professionals. After Data synthesis and treatment, authors have defined Three main axes of the subject which are: 1. The role and impact of HIT on care delivery during the present COVID-19 crisis and The breadth of HIT in the health care sector during the pandemic from an international perspective 2. How developing countries, such as Morocco, have responded to Covid 19 by overhauling and modernizing their hospital information systems at the operational, tactical, and strategic levels 3. Morocco’s Accelerated Path to Digital Health Transformation. Finally, the authors presume that hospitals should consider improved data system integration, the use of customized warnings, and the growth of telehealth while employing information and communication technology to offer efficient medical care. In the Moroccan context, it was an excellent chance for health professionals to place greater emphasis on the health information system and to comprehend the critical role that information technology played throughout the pandemic era. Keywords: Healthcare · Information technology · Covid 19 · Healthcare information sys- tems · Patient care · Patient information · Medical Care · EHR © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 46–57, 2023. https://doi.org/10.1007/978-3-031-35248-5_5
How Health Information Technology Improved Patient Care and Treatment
47
1 Introduction The name Covid-19 refers to a new kind of coronavirus that is raising widespread worry throughout the world. The corona virus, which was identified in the Chinese city of Wuhan, is a kind of virus known to cause respiratory tract diseases, according to the Centers for Disease Control and Prevention, cdc.gov. On 30 January 2020, nearly two months after the epidemic began, the World Health Organization (WHO) declared a worldwide emergency against the coronavirus, noting the infection’s spread to 18 countries [1]. This paper will discuss the role of technology and informatics in responding to the Covid-19 epidemic in depth. The writing technique includes of a meta-synthesis of prior publications and research on IT tools and systems utilized by various nations to handle the Covid-19 epidemic from multiple journal database sources. This report should be one of the comprehensive infor- mation sources summarizing the role of information technology in the effective handling of the Covid-19 epidemic. The study’s goal is to learn and show the important role of healthcare information technology during the COVID-19 (coronavirus 2019), and how countries like Morocco can benefit from an international case study to reorganize and revolutionize their hospital information systems.
2 Background The literal definition of health information technology (HIT) is the applicability of information based technologies, methods, applications, and so on in health care-related services[b]. In a wider perspective, information and communication technologies (IT) are strategies that use cloud-based databases to keep records of patient information, reports, etc. in order to manage all situations properly and efficiently [2, 3]. HIT enables medical workers to manage a plethora of operations, such as prescriptions for patients, the production of electronic health records, testing and analysis data, and so on [4–6]; [7]. It also generates and promotes a quality environment in which technology and healthcare may interact with one another to improve pa- tient care and happiness [8–12]. In the current COVID-19 epidemic, where medical consultation has become difficult for every- one, especially patients, HIT has emerged as a critical game-changer by ordering its extended services to physicians, healthcare patients, and so on [13, 14]. HIT intends to automate the control and monitoring of patients’ health via a networked channel of intelligent information-based applications [15, 15, 17, 18]. The primary benefits of using the HIT concept in the COVID-19 pandemic include record confidentiality, level of care, interactive consultation, proper communication, and so on [19–21]. The Fig. 1 depicts all of the data management classes offered at COVID-19 for those interested in using HIT in healthcare.
48
A. Kadiri et al.
HIT for Healthcare in COVID-19
It includes all the personal health information
Personnnel Health Records
It includes records from the healthcare providers Electronic Health Records
It includes only the clinical records Electronic Medical Records
Fig. 1. Data storage and handling in HIT for healthcare during COVID-19
The proposed approach for information and communication technologies is generally concentrated on the use of information technologies in the sphere of health care and medicine. It all began in the 1920s with the publication of the first medical records. Subsequent shifts occurred in the 1960s and 1970s, when computer technology drew the attention of health care professionals by allowing them to store and exchange medical data and information. The main spike happened in 2004, when United States President George W. Bush formed a National Coordinating Office for Information and Communication Technology (IT) with the purpose of unifying IT across the country [22–24]. The Fig. 2 shows the appropriate planned uses of HIT in healthcare during the COVID-19 pandemic.
How Health Information Technology Improved Patient Care and Treatment
49
Digital Systematic Prescription
Tele/Cloud Based Healthcare
Healthcare Decision Support
Proposed Application of HIT in healthcare during COVID-19
Electronic Information System
Computerized System Network
Advanced Case Handling
Fig. 2. Cycle view of proposed services through HIT in healthcare
Science, engineering, and technological advances have spanned multiple decades and continue to this day. The primary objective for this is to assist humanity in its daily lives and times of need. Information and communication technology (ICT) tools, procedures, software, and so on must be made available to patients in order for them to be supported and treated [25]. Information and communication technologies employ complex network-based tactics to provide more effective services to medical personnel, health care institutions, patients, and others [26–28]. Figure 3 depicts the healthcare benefits and assistance provided during COVID-19.
50
A. Kadiri et al.
Reductionin paper Work
Effective Work Flow Process
Superior Healthcare Quality
HIT healthc are during COVID-19
Accurate & Error free Facilities
Highly Productive Care
Services at Low Cost
Fig. 3. Proposed enhancements through HIT for healthcare in the COVID-19 period
3 Research Method This section will review the synthesis results of previous articles and research to answer the following research questions: • The role and impact of HIT on care delivery during the present COVID-19 crisis and The breadth of HIT in the health care sector during the pandemic from an international perspective • How developing countries, such as Morocco, have responded to Covid 19 by overhauling and modernizing their hospital information systems at the operational, tactical, and strategic levels. • Morocco’s Accelerated Path to Digital Health Transformation 3.1 The Role and Impact of HIT on Care Delivery During the Present COVID-19 Crisis and the Breadth of HIT in the Health Care Sector During the Pandemic from an International Perspective In this section, we refer to studies that focused on gaining an in-depth insight into how hospitals with a deep and rich experience using health information technology (IT) handled the COVID- 19 (coronavirus 2019) pandemic from an IT point of view. In [29], the study was based on interviews with 44 health professionals with informatics training
How Health Information Technology Improved Patient Care and Treatment
51
in 6 international hospitals. A thematic approach was used to develop a framework for coding and to identify potential and emerging topics. The figure below gives more details on the site and the participants who were involved in this study: Table 1. Participants by site involved in the study. Site Hospital Details Identifier Location Size Type
Participant Details Roles Included in Sample
Total
COVID-19Related Optimizations
Site1
United 760 Kingdom beds
Teaching Pharmacy managers, 6 hospital analysts,pharmacists, nurses, information officers
Combining data sources, CDS, alerts,telehealth, remote ward rounds
Site2
United 800 Kingdom beds
Teaching Pharmacy managers, hospital physicians, analysts, pharmacist, Nurses Other Ancillary care
13
Expediting the transition of paper-based processes to control contagion and to allow remote working
Site3
United States
800 beds
Teaching Pharmacy managers, hospital physicians, analysts, pharmacists
8
Integration and interoperability of systems, CDS, modified alerts, telehealth
Site4
United States
670 beds
Teaching Physicians, nurses hospital
3
CDS, EHR assisted infection control and discharge
Site5
United States
1500 Teaching Pharmacy managers, beds hospital physicians, pharmacists Information officers
5
CDS, integration and interoperability of systems, modified alerts, remote working,telehealth
Site6
United States
80 beds
9
Alerts to inform physicians of COVID/testing status of patients. Adding PCR to the system
Pediatric cancer analysts
Pharmacy managers, physicians, information officers
52
A. Kadiri et al.
BoB: best of breed; CDS: clinical decision support; COVID-19: coronavirus disease 2019; EHR: electronic health record; PCR: polymerise chain reaction. As a result of [29], the authors have listed the following main topics: • Time and resources were administered through information and communication technologies • All sites agreed that improved healthcare system integration was crucial. • Generating alerts using data from several data sources improved processing. • Emergency stop warnings and standard operating procedures were shown to be useful for selecting appropriate therapeutic remedies in the event of anticipated drug shortages. From [29], during the COVID-19 pandemic, leading patient safety and medical informatics spe- cialists worked to enhance their respective site EHR systems, and this report shares their per- spectives and experiences. Not unexpectedly, many systems underwent significant and quick adjustments in order to adapt, which were aided by reduced governance and the employment of IT experts to enhance processes. The utility of current EHR systems to enable the fast spread of telehealth, the relevance of in- teroperability and the availability of patient data to the provider, and resource conservation methods were cited as key points in [29]. While some of these lessons have already been learnt and implemented in technologically modern hospitals, this study may lead to viable tactics for less digitally mature institutions in a context of developing countries such as Morocco, as well as responses to future pandemics of this type. Aside from the numerous roles and applications of the HIT concept, there are some challenges to overcome while adopting this approach to treat patients during COVID19. These difficulties, including global/local adoption of an interoperability culture, are contingent on physician and hospital cooperation to exchange transferred/referred patient records, data, and so on. It is always difficult to incorporate HIT facilities in a smart and connected healthcare network [30–32]. Data safety and security are two more areas where effective protecting for highly confidential information communicated over the cloud-based route must be designed. 3.2 How Developing Countries, Such as Morocco, Have Responded to Covid 19 by Over- Hauling and Modernizing Their Hospital Information Systems at the Operational, Tac- Tical, and Strategic Levels. Many developing countries, like Morocco, are grappling with the difficulty of providing inex- pensive, accessible, egalitarian, and high-quality health care [33]. Morocco’s health statistics have been steadily improving, although there is always a real need for improving healthcare quality of services [34]. Telemedicine was developed with the intention of utilizing contemporary information and communication technologies [ICTs] to allow as many citizens as possible to swiftly and conveniently obtain high-quality treatment from any location and at any time among medical experts [35]. Goals for sustainability have emerged as a global strategy for solving significant global concerns [36]. Digital
How Health Information Technology Improved Patient Care and Treatment
53
health has been presented as a way to provide better and more widespread healthcare in society, and its adoption has the potential to contribute to the attainment of SDG 3 [37]. The pandemic’s growth in Morocco has demonstrated a regulated tendency throughout the curfew period in Morocco from March to May 2020 [38, 39], 40]. The epidemiological indicators favored a stepwise zone-by-zone deconfinement process, which began on June 10, 2020, after three months of severe confinement. However, the pandemic quickly accelerated after this process began, increasing disease occurrence and dissemination. With the rate of people impacted (more than 500.000) following an uncontrollable curve, an obligation to control the quality of health care service through increased resources to meet a very urgent need is present, and the only way to achieve this is through a fairly robust healthcare digitization process [41]. Moroccan digital health professionals may rely on a competent digital infrastructure and information framework [42] despite electricity constraints in rural areas, IT infrastructure to be improved and cultural barriers [43, 44]. In 2017, Morocco was rated fourth [45] among African countries for both internet and mobile connectivity, which includes fiber optic, 3G, 4G, and continuing 5G deployment [46]. Furthermore, the electrification problem is continuously improving [47]. A culture of digitalization of the various processes on the professional and patient side has been developed in the health sector as a result of its effectiveness in treating patients and preventing the risk of the pandemic. Morocco’s current ranking in the “Digital Riser Report of 2020” [48, 49] attests to the Moroccan government’s dedication, drive, and support in achieving a Moroccan’s healthcare digitization Project. International countries focused towards well health care strategies and approaches such as proactive surveillance, infection prevention and control, confinement, and therapeutic interventions, in addition to a rational vaccination strategy, throughout the covid-19 era [50]. The countries that successfully managed the epidemic appear to have implemented new technologies and health surveillance in their health systems [42, 51]. Morocco has used since the beginning of the pandemic an integrated digital health approach to control and mitigate its consequences based on several digitalization work streams. Below are the different strategic axes on which Morocco has focused: • Telemedicine: A remote medical practice using new information and communication technologies. Providing access to medical care for all throughout the country through mobile phone apps, telemedicine platforms and so on. • Epidemiological monitoring: Collection and analysis of health related events on a regular basis. • Contact tracking: Was insured through Wiqaytna mobile app. • Smart vaccination: Vaccination campaign launched via text messages, mobile application and website platforms. 3.3 Morocco’s Accelerated Path to Digital Health Transformation As harmful as the COVID-19 epidemic has been for international health systems, it has also played a vital role in Morocco’s digital health revolution by bringing attention to the need for digitization in the healthcare sector among practitioners and decision-makers.
54
A. Kadiri et al.
In this perspective, Morocco has launched a strategic plan to improve the performance of the health system in terms of standardization of processes, reorganization taking into consideration the need to have a system that facilitates the link between health personnel and patients [53]. The main focus regarding the process of healthcare improvement and digitization, based on the Moroccan Ministry of Health 2025 healthcare roadmap is [53]: • Implement the Ministry of Health’s IT master plan. • Complete and make it successful the deployment of the decision-making information system • Integrate the private sector into the Ministry of health information system. • To design, computerize and implement the “shared medical record”. • To create the Data Management Center of expertise of the Ministry of Health. • To implement an information system security policy. • Computerize the drug supply cycle and digitize the distribution their distribution logistics system. The digital transition is a bold strategy for accomplishing these objectives, making it the most comprehensive Moroccan healthcare-IT strategy yet. it strengthens the alreadyexisting IT infrastructure [54].
4 Conclusion In the context of the COVID-19 pandemic, hospitals worldwide should explore enhanced data system integration, the use of personalized alerts, and the expansion of telehealth while applying information and communication technology to provide medical treatment. It was an opportune chance for Moroccan health professionals to place a higher focus on the health information system and understand the crucial role that information technology played during the epidemic era and to and to consider strategic technical and operational plans for the development of their IT infrastructure.
References 1. Singh, R.P., Javaid, M., Haleem, A., Vaishya, R., Bahl, S.: Significance of health information technology [HIT] in context to COVID-19 pandemic: potential roles and challenges. J. Ind. Integr. Manage. 18, 100139 (2020) 2. Wei, L.L.Y., Ibrahim, A.A.A., Nisar, K., Ismail, Z.I.A., Welch, I.: Survey on geographic visual display techniques in epidemiology: Taxonomy and characterisation. J. Ind. Inf. Integr. 18, 100139 (2020) 3. Tan, T.K., Gudi´c, M., Flynn, P.M., (eds). Struggles and Successes in the Pursuit of Sustainable Development. London, Routledge, p. 238 (2020) 4. Li, S., Xu, L.D., Wang, X.: A continuous biomedical signal acquisition system based on compressed sensing in body sensor networks. IEEE Trans. Ind. Inf. 9(3), 1764–1771 (2013) 5. Xu, L.D., Xu, E.L., Li, L.: Industry 4.0: state of the art and future trends. Int. J. Prod. Res. 56(8), 2941–2962 (2018) 6. Srinivasa, K.G., Anupindi, S., Kumar, A.: Analytics on medical records collected from a distributed system deployed in the Indian rural demographic. J. Manage. Anal. 5(1), 54–72 (2018)
How Health Information Technology Improved Patient Care and Treatment
55
7. Aceto, G., Persico, V., Pescape, A.: Industry 4.0 and health: Internet of things, big data, and cloud computing for healthcare 4.0. J. Ind. Inf. Integr. 18, 100129 (2020) 8. Deimazar, G., Kahouei, M., Zamani, A., Ganji, Z.: Health information technology in ambulatory care in a developing country. Electron. Physician 10(2), 6319 (2018) 9. Xu, L.D., Duan, L.: Big data for cyber physical systems in industry 4.0: a survey. Enterp. Inf. Syst. 13(2), 148–169 (2019) 10. Park, S., Choi, G.J., Ko, H.: Information technology-based tracing strategy in response to COVID-19 in South Korea — privacy controversies. JAMA 323(21), 21279–22130 (2020) 11. Kogan, M., Klein, S.E., Hannon, C.P., Nolte, M.T.: Orthopaedic education during the COVID19 pandemic. J. Am. Acad. Orthop. Surg. 28, e456–e464 (2020) 12. Shen, W., Yang, C., Gao, L.: Address business crisis caused by COVID-19 with collaborative intelligent manufacturing technologies. IET Collaborative Intell. Manufact. 2(2), 96–99 (2020) 13. Khan, I.H., Javaid, M.: Big data applications in medical Field: a literature review. J. Ind. Integr. Manage. 5(3), 1–17 (2020) 14. Delen, D.N., Dorokhov, O., Dorokhova, L., Dinçer, H., Yüksel, S.: Balanced scorecard based analysis of customer expectations for cosmetology services: a hybrid decision modeling approach. J. Manage. Analytics 7(5), 1–32 (2020) 15. Da Xu, L., Wang, C., Bi, Z., Yu, J.: Object-oriented templates for automated assemblyplanning of complex products. IEEE Trans. Autom. Sci. Eng. 11(2), 492–503 (2013) 16. Da, X.: L. A Systems Perspective on Industrial Information Integration. CRC Press, Enterprise Integration and Information Architecture (2014) 17. Xu, B., Xu, L.D., Cai, H., Xie, C., Hu, J., Bu, F.: Ubiquitous data accessing method in IoT based information system for emergency medical services. IEEE Trans. Ind. Inf. 10(2), 1578–1586 (2014) 18. Liu, J., Kang, N., Man, Y.: Evidence fusion theory in healthcare. J. Manage. Analytics 5(4), 276–286 (2018) 19. Xu, L.: Systems science systems approach and information systems research. Syst. Res. Behav. Sci. 17(2), 103–104 (2000a) 20. Richardson, J.E., Abramson, E.L., Pfoh, E.R., Kaushal, R., HITEC Investigators.: How communities are leveraging the health information technology workforce to implement electronic health records. In: AMIA Annual Symposium Proceedings, pp. 1186–1195 (2011) 21. Hernandez-Orallo, E., Manzoni, P., Calafate, C.T., Cano, J.C.: Evaluating how smartphone contact tracing technology can reduce the spread of infectious diseases: The case of COVID19. IEEE Access 8, 99083–99097 (2020) 22. Liang, J., et al.: Evaluating the applications of health information technologies in China during the past 11 years: Consecutive survey data analysis. JMIR Med. Inform. 8(2), e17006 (2020) 23. Wang, G., Zhang, Y., Zhao, J., Zhang, J., Jiang, F.: Mitigate the e®ects of home confinement on children during the COVID-19 outbreak. Lancet 395(10228), 945–947 (2020) 24. Roy, S., et al.: Deep learning for classification and localisation of COVID-19 markers in point-of-care lung ultrasound. IEEE Trans. Med. Imaging 39(8), 2676–2687 (2020) 25. Suman, R., Javaid, M., Haleem, A., Vaishya, R., Bahl, S., Nandan, D.: Sustainability of corona- virus on di®erent surfaces. J. Clin. Exp. Hepatol. 10(4), 386–390 (2020) 26. Qin, C., et al.: Dysregulation of immune response in patients with COVID-19 in Wuhan China. Clin. Infect. Dis. 71(15), 762–768 (2020) 27. Javaid, M., Haleem, A.: Critical components of Industry 5.0 towards a successful adoption in the field of manufacturing. J. Ind. Integr. Manage. 5, 327–348 (2020) 28. Lite, S., Gordon, W.J., Stern, A.D.: Association of the meaningful use electronic health record incentive program with health information technology venture capital funding. JAMA Network Open 3(3), e201402–e201402 (2020)
56
A. Kadiri et al.
29. Malden, S., Heeney, C., Bates, D.W., Sheikh, A.:J. Am. Med. Inf. Assoc. 28(7), 1555–1563 (2021) 30. Hemmat, M., Ayatollahi, H., Maleki, M.R., Saghafi, F.: Future research in health information technology: a review. Perspectives in Health Information Management, PMCID: PMC5430110 (2017) 31. Rioth, M.J., Warner, J., Savani, B.N., Jagasia, M.: Next-generation long-term transplant clinics: improving resource utilisation and the quality of care through health information technology. Bone Marrow Transplant. 51(1), 34–40 (2016) 32. Guo, X., et al.:Survey of COVID-19 disease among orthopaedic surgeons in Wuhan, People’s Republic of China. J. Bone Joint Surg. Am. 102, 847–854 (2020) 33. Tan, T.K., Gudi´c, M., Flynn, P.M., (eds.) Struggles and Successes in the Pursuit of Sustainable Development. London, Routledge, p. 238 (2020). https://doi.org/10.4324/9781351140560 34. Lupton, D.: Critical perspectives on digital health technologies. Sociol. Compass. 8, 1344– 1359 (2014). https://doi.org/10.1111/soc4.12226 35. Latifi, R.: Current Principles and Practices of Telemedicine and E-health. Amsterdam; Oxford, IOS Press, p. 297 (2008) 36. Pradhan, P., Costa, L., Rybski, D., Lucht, W., Kropp, J.P.: A systematic study of sustainable development goal [SDG] interactions. Earth’s Future 5, 1169–1179 (2017). https://doi.org/ 10.1002/2017EF000632 37. Asi, Y.M., Williams, C.: The role of digital health in making progress toward Sustainable Development Goal [SDG] 3 in conflict-affected populations. Int. J. Med. Informat. 114, 114– 20 (2018). https://doi.org/10.1016/j.ijmedinf.2017.11.003 38. Maneesh, P., El Alaoui, A.: How Countries of South Mitigate COVID-19: Models of Morocco and Kerala, India. Rochester, NY: Social Science Research Network (2020) 39. Alami, A.E., Fattah, A., Chait, A.: Medicinal plants used for the prevention purposes during the Covid-pandemic in Morocco. J. Anal. Sci. Appl. Biotechnol. 1, 4–11 (2020) 40. Worldometer. Morocco COVID. Available online at: https://www.worldometers.info/coronavirus/country/morocco/ 41. Balta, M., Valsecchi, R., Papadopoulos, T., Bourne, D.J.: Digitalization and co-creation of healthcare value: a case study in Occupational Health. Technol Forecast Soc Change. 168, 120785 (2021). https://doi.org/10.1016/j.techfore.2021.120785 42. haskar, S., Bradley, S., Chattu, V.K., Adisesh, A., Nurtazina, A., Kyrykbayeva, S., et al.: Telemedicine across the globe-position paper from the COVID-19 pandemic health system resilience program [REPRO- GRAM ] International Consortium [Part 1]. Front Public Health. vol. 8, pp. 556720 (2020). https://doi.org/10.3389/fpubh.2020.556720 43. Borketey, P.: A Literature Review on the Challenges of eHealth Implementation in Developing Countries Among Rural Folks: A Case of Ghana. Yrkeshögskolan Arcada (2017). http://www. theseus.fi/handle/10024/139551 44. Garga, H., Evina, C.D., Vouking, M., Tamo, V.C.: Are e-health programs effective in LMIC? SURE Rapid Response (2012). http://www.cdbph.org/documents/Rapid_Response_Effective ness_of_e_health_programs_in_LMIC_july2012.pdf 45. Ecofin, A.: Les 46 Pays Africains avec la plus Grande Vitesse De Téléchargement [Infographie]. https://www.agenceecofin.com/telecom/0903-74625-les-46-pays-africa ins-avec-la-plus-grande-vitesse-de-telechargement-infographie 46. Ecofin, A.: Classement Des Pays Africains en Matière de Connectivité Mobile, Selon la GSMA. https://www.agenceecofin.com/gestion-publique/0908-49318-classement-des-paysafricains-en-matiere-de-connectivite-mobile-selon-la-gsma. Accessed 17 Apr 2022 47. African Development Bank Group. Bank AD. Morocco Making Giant Strides in Electrification. African Development Bank - Building today, a better Africa tomorrow (2019). https:// www.afdb.org/en/news-and-events/morocco-making-giant-strides-in-electrification-18376
How Health Information Technology Improved Patient Care and Treatment
57
48. Agence de Développement du Digital. Présentation de l’ADD. https://www.add.gov.ma/pre sentation-de-ladd 49. European Center for Digital Competitiveness. https://digital-competitiveness.eu/ 50. Whitelaw, S., Mamas, M.A., Topol, E., Spall, H.G.C.V.: Applications of digital technology in COVID-19 pandemic planning and response. Lancet Digital Health. 2, e435–e440 (2020). https://doi.org/10.1016/S2589-7500[20]30142-4 51. Bhaskar, S., Bradley, S., Chattu, V.K., Adisesh, A., Nurtazina, A., Kyrykbayeva, S., et al.: Telemedicine as the new outpatient clinic gone digital: position paper from the pandemic health system REsilience PROGRAM International Consortium [Part 2]. Front Public Health. 8, 410 (2020). https://doi.org/10.3389/fpubh.2020.00410 52. Ting, D.S.W., Carin, L., Dzau, V., Wong, T.Y.: Digital technology and COVID-19. Nat Med. 26, 459– 61 (2020). https://doi.org/10.1038/s41591-020-0824-5 53. Moroccan Ministry of Health. PLAN SANTÉ 2025 . http://www.draatafila-let.ma/ima ges/Publications-pdf/Plan-de-sant%C3%A9-2025.pdf 54. Parks, R., Wigand, R.T., Othmani, M.B., Serhier, Z., Bouhaddou, O.: Electronic health records implementation in Morocco: Challenges of silo efforts and recommendations for improvements. Int. J. Med. Inf. 129, 430–437 (2019)
Dates Detection System for Automatic Harvesting Using Deep Learning Yousra Zarouit1,2(B) , Brahim Aksasse1,2 , and Mohamed Ouhda3 1 Department of Computer Science, Faculty of Sciences and Technics, Moulay Ismail
University, Errachidia, Morocco [email protected], [email protected] 2 MAIS Laboratory, Department of Computer Science, Faculty of Sciences, Moulay Ismail University Meknes, Meknes, Morocco 3 Department of Computer and Mathematics, Higher School of Technology, Sultan Moulay Slimane University, Khenifra, Morocco
Abstract. An accurate vision system for detecting and analyzing fruit in realtime requires fast and accurate algorithms for object detection. This will allows robotic systems to analyze and process in real-time identically like to the human visual system making them more efficient and cost-effective. However, success in this domain is still limited. No research has addressed the problem of date fruit detection and localization in an orchard environment. The system of our interest is a date fruit-harvesting robot. The robot will detect, analyze and cut the date bunch from the palm tree. In this work, a state of art object detection algorithm, the You Only Look Once version 3 (YOLOv3), is used to solve the detection and localization of date bunches problem. Unstructured agricultural environments with constantly changing light, and shadow, make date fruit detection challenging. Experimental results show the effectiveness of the proposed algorithm to deal with these problems, giving high accuracy. Based on the model output, the robot can detect and locate dates in images to make the harvesting decision. Keyword: Dates fruit detection · Automated harvesting · Deep learning · Convolutional neural networks · YOLOv3
1 Introduction The date palm (Phoenix dactylifera) is one of the oldest cultivated plant species and is potentially the oldest domesticated tree in the world. Palm is a tree of great interest because of its high productivity and nutrition quality. Date combines the sweet taste and health benefits it is a rich source of minerals, dietary, fiber, carbohydrates, sugar, and vitamins and its adaptability to Saharan regions. In addition to its ecological and social roles, the date palm contributes mainly to the agricultural income of the peasants, offers dates and a multitude of by-products for domestic, artisanal, and industrial use. According to the Food and Agriculture Organization, the global production of date is 8.5 million tons, the Top fifteen date producer countries in 2018 [1] are depicted in (Fig. 1). That makes dates one of the main economic pillars for those countries. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 58–74, 2023. https://doi.org/10.1007/978-3-031-35248-5_6
Dates Detection System for Automatic Harvesting Using Deep Learning
59
Fig. 1. Top 13 Date Producing Countries According to Food and Agriculture Organization of the United Nations, FAOSTAT (2018).
For example, in Morocco and according to the Ministry of Agriculture, Fisheries and International Exhibition of Dates MAPM, SID [2] (Fig. 2), the agricultural area dedicated to dates in morocco increased from 48,000 hectares in 2010 to 59,640 hectares in 2018. The production reached 112,000 Ton in 2017–2018. On the other side, the number of feet has evolved from 4.8 million in 2010 to 6.9 million in 2017–2018. Date production in Egypt was around 541,963 tons in 1990, approximately reaching 1,562,171 tons in 2018. Consequently, Egypt is the first largest date-producing country in 2018 and the second in 1990. (Table 1) illustrates the top five date-producing countries in 1990 and 2018.
Fig. 2. Evolution of date production in Morocco Source: MAPM, SID
60
Y. Zarouit et al. Table 1. Top five date-producing countries in 1990 and 2018.
Country
2018
1990
Rank
Quantity (ton)
Order
Quantity (ton)
Egypt
1
1,562,171
2
541,963
Saudi Arabia
2
1,302,859
3
527,881
Iran
3
1,204,158
4
516,295
Algeria
4
1,094,700
6
205,907
Iraq
5
614,584
1
544,930
Although there are more than 400 varieties of dates that are described differently by shapes, texture, color, and flavors, their nutritional value is still the same. Dates have four essential stages of maturity; known as Immature (Kimri), Bisdr (Khalal), Rutab, and Tamer stage (Fig. 3). The harvesting decision depends on Varieties type, climatological situations, and market demands.
Fig. 3. The four dates fruit maturity stages
Dates production grows year per year that makes harvesting difficult. However, most techniques used for date harvesting have not evolved, still being handled manually, which causes waste of time and is the source of delay in the production cycle of dates, necessitating a change into automating. Note that automatic harvesting can increase the quality and yield as well as reduce the production time, costs and avoid work accidents that would make the product more competitive. The most advantage of harvesting robots is their ability to analyze, classify and decide in real-time. Indeed, the coupling of visual data with machine vision algorithms makes numerous operations easier and more efficient, gainful and advances agricultural automation to new levels. Unfortunately, success in this area is still limited due to several challenges caused by many objects of different colors, shapes, sizes, textures, and reflective properties. Unstructured agricultural environments with a high degree of uncertainty, constantly changing lighting; and shadow conditions [3]. Hence, in this research, we attempt to detect and classify different types of dates using computer vision in the natural environment. Many researchers have provided different algorithms for detection, localization, and harvest for objects other than dates such as apples [4], sweet pepper [5], tomato [6], and
Dates Detection System for Automatic Harvesting Using Deep Learning
61
strawberry [7]. Nowadays on the other hand, research on automated date fruit harvesting stays very limited. Most of the previous researchers focused on post-harvesting, we can divide them into three principal categories (quality and gradient, maturity analysis, variety classification) due to the importance of dates, most of the previous researches concerning inspection and grading of date fruit quality. We can evaluate the quality of dates by several factors such as surface defects, sugar content, and hardness. In terms of image analysis, the problem can be seen as a classification problem in machine learning. For inspection and grading [8, 9] proposed two techniques using co-occurrence matrix and color machine vision technique. In [10], the authors use three classifiers k-nearest Neighbors (KNN), Support Vector Machine (SVM), and Linear Discriminant Analysis (LDA), to grade the dates into six grades. Otherwise, [11] hardness is one of many factors considered for the quality of dates the authors use stepwise and linear discriminate analysis to classify dates into three grades soft, semi-hard, and hard. To classify dates according to their type, [12] propose a method based on fifteen features extraction including (color, size, shape, and texture) to classify date fruit into seven varieties using 140 images in the experiment. The study in [13] uses shape, size, texture, local descriptor, and support vector machine (SVM) with Radial Basis Function (RBF) kernel to classify four types of dates to give better accuracy of 98.1%. [14] has proposed a single date classifier based on the Gaussian mixture model and four shape features including area and eccentricity using 5000 images of ten date varieties. [15] developed a classifier to classify five date varieties using color properties in RGB space and probabilistic neural networks. Moreover, all the previous researches use images for individual dates after harvesting taken from a close distance. Hence these approaches were solving simple problems and are not robust enough to work in the natural environment. In the most recent research, a case study [16] showed that the automatic classification of date fruit in smart cities using deep learning techniques with fine-tuning and pre-trained models, in addition to fog and cloud computing. [17] proposed date fruit classification models to classify dates according to their types, maturity, and the harvesting decision. Note that none of these previous works has investigated the problem of detecting date fruits in an orchard environment. They used some classification techniques to classify the entire image. Unfortunately, they did not locate the exact location of dates in the input image, which is the core information for the robot to harvest dates. To realize an automatic robot for harvesting biggest challenges facing machine vision tasks caused by unstructured environments, fruit by different size, shape, color, and texture. Moreover, for constantly changing lighting and shadow conditions, the date palm is different and more complicated than other trees due to several conditions. Firstly, the height of the date palm tree varies and it can reach up to 23 m in height. Secondly, there are more than 400 varieties of date closely very similar making their classification a challenging task. Thirdly, there are different harvesting methods used, the single brunch content individual dates in a different stage of maturity, so the technique of harvesting can be by cutting the whole branch or by picking individually dates one by one. The rest of the paper is structured as follows: Sect. 2 introduces the materials and methods used in this research on the detection of dates. Section 3 presents the proposed
62
Y. Zarouit et al.
method. In Sect. 4, the experimental setup and the results obtained for the proposed solution. Finally, Sect. 5 provides some conclusions and future works.
2 The Problem Formulation To realize a date fruit-harvesting robot, we need to detect and locate dates in the palm. For this task, instead of using traditional image processing techniques, we will use the deep learning methods more exactly Convolutional Neural Networks (CNNs) to detect dates from the images because of their high accuracy. 2.1 CNN Deep Learning Deep learning is a type of artificial intelligence derived from machine learning where the machine can itself learn only from examples without being explicitly programed. Deep-learning becomes the most used approach applied in many domains like object detection, image classification, Content-Based Image Retrieval [18] video recognition, classification, speech recognition and Natural Language Processing, etc. CNN is a sub-category of neural networks, it is considered one of the most successful image classification models. CNN uses a system like a multilayer perceptron but, with the advantage of reducing the number of calculations, the structure of CNN is composed of input layers, a hidden layer made up of many convolutional layers, pooling layers, fully connected layers, and normalization, then the output layers as given in the following (Fig. 4).
Fig. 4. An example of CNN architecture
To train the CNN model from scratch, we need a large dataset to learn millions of parameters. There are various datasets that we can leverage for applying convolutional neural networks like ImageNet [19]. But in our case, classifying dates fruit it’s difficult due to the need for a large labeled dataset, and it will take a large number of resources (time and computation power) to train big models from scratch. On the other hand, transfer learning comes to resolve those problems using the knowledge acquired by a neural
Dates Detection System for Automatic Harvesting Using Deep Learning
63
network during the resolution of other cases to improve the learning rate in the target task. By using transfer learning, we improved baseline performance and improvements in the overall time taken to learn a model. Another section of computer vision is object detection, which is applied in many tasks like pose estimation, surveillance, vehicle detection, etc. There is a difference between object detection algorithms and classification algorithms. Classification algorithms consist of recognizing to which class an image belongs among a set of classes predetermined. Contrariwise, object detection is an image recognition technology used to locate one or more objects in an image or video by drawing a bounding box around it. 2.2 Object Detection Object detection is a computer vision task and image processing technology that deals with detecting objects of different classes (such as animals, humans, or plans) in videos and images. Object detection becomes one of the most research areas that have received a great deal of attention in recent years due to its importance and challenging problems in computer vision. The main objective of object detection is to develop computational models that respond to a principal question in the computer version (What objects are where?). In recent years different object detection models have been developed and used in research on fruit detection, such as Region-based Convolutional Neural Networks(RCNN) [20], RetinaNet [21], and YOLO [22]. Note that object detection algorithms can be divided into two categories namely: twostage detectors and one-stage detectors. The first one uses at the first step Region Proposal Network to generate regions of interest, then sends these region proposals to the pipeline for object classification task and bounding-box regression. The experimental of these networks produce higher accuracy rates but are usually slower. R-CNN and Faster RCNN [23] are networks belonging to this category. On the other side, one-stage detector algorithms manipulate object detection as a regression problem by taking an input image and learning the class probabilities, and bounding box coordinates simultaneously. In terms of accuracy, one-stage detector algorithms have lower accuracy but are much faster than two-stage detectors. YOLO (You Only Look Once), RetinaNet, and Single Shot MultiBox Detector (SSD) [24], are included in this category. 2.3 Yolo You Only Look Once (YOLO) is an object detection model proposed by R. Joseph et al. in 2015. It is an object detection algorithm much different from region-based algorithms and it is the first one-stage detector in deep learning where a single convolutional network divides the image into regions then predicts the bounding boxes and class probabilities for each Box. YOLO highly fast, fast version of YOLO attains 155 fps (Frame Per Second) with VOC 2007 (The PASCAL Visual Object Classes Challenge 2007) mAp = 52.7% (mean Average Precision), while its enhanced versions run at 65 fps on Tesla V100 GPU with 43.5% AP (65.7% AP50) for the MS COCO dataset. The YOLO-V3 [25] network evolved from the YOLO, YOLO-V2, which improves detection accuracy and detection speed. Compared to other classic object detection models such as the RCNN series YOLO eliminates the use of a large number of sliding windows (anchor)
64
Y. Zarouit et al.
and regional proposals and generates class probabilities and bounding box coordinates directly through regression, which improves the real-time performance of the target detection. YOLO-V3 uses as a feature extractor Darknet [26] architecture, which consists of 53 layers, trained with the ImageNet dataset, to detect object an additional 53 layers trained with the Pascal VOC dataset were adding.YOLO-V3 surpasses most of the detection algorithms, using residual connections and upsampling. The model can perform detection at three different scales in specific layers making YOLO-V3 more efficient to detect small objects. But, due to the complexity of the solution processing time becoming slower than the previous versions, (Fig. 5), shows the structure of YOLO-V3.
Fig. 5. Simplified layer architecture of YOLOv3 network
YOLO divides the image into S*S grids. For each grid, run a CNN predicting B Bounding boxes and confidence scores for each Box. The Bounding boxes contain five predictions (dx, dy, dw, dh, pc) where pc is the probability of detecting an object and (dx, dy, dw, dh) are the properties of the detected bounded area. For each grid, cpi is a binary representation of one of the detected classes. In the final steps, run the Non-maximum Suppression to remove potential duplicates that overlap bounding areas. In sum, the loss function defined as: 2 2 s2 B w w obj + λcoord d w − dˆ d w − dˆ i
ij
i
i=0 j=0 2
+
s B
obj p p 2 ci − cˆ i
ij
i=0 j=0 2
+ λnoobj
s B i=0 j=0
noobj p p 2 ci − cˆ i
ij
i
i
Dates Detection System for Automatic Harvesting Using Deep Learning 2
+
s i=0
p 2 pi ci − pˆ i cip
obj
1ij
65
(1)
c∈classes
where λcoord increases the weight of the localization loss usually = 5, s2 present the output feature map of all grid cells, B : number of the bounding box for each grid cell, obj noobj obj : the complement of ij , c ij indicate if an object appears in the cell i and ij p p real object confidence, c predicted object confidence, pi ci , pˆ i ci : the probability of real and predicted category.
3 The Proposed Solution The proposed system consists of an object detection algorithm to locate date fruit in the palm. The system input receives a stream of images or frames from the camera then the system output is date detection with localization. The model uses transfer learning with fine-tuning based on a pre-trained CNN model to extract features. In this proposed method, we use the YOLO-V3 algorithm to locate dates on the palm also DARKNET53 as a backbone. The block diagram in (Fig. 6) provides an overview of the proposed framework. 3.1 Dataset One of the biggest problems to build a robust vision system is the data, more precisely in our case, a database of date fruit images taken in a natural orchard under several natural conditions. Fortunately, there is a dataset built-in 2019 by the Center of Smart Robotics Research (www.CS2R.ksu.edu.sa) called “DATE FRUIT DATASET FOR AUTOMATED HARVESTING AND VISUAL YIELD ESTIMATION” [27], the database divided into two different datasets. The first one contains 8079 images of 29 date palm bunches belong to five types (Meneifi, Naboot Saif, Sullaj, Khalas, and Barhi) captured at different stages of maturity (immature, Khalal, Rutab, and Tamer). Dates images include different angles and scales, dates bunches covered by bags, and various daylight conditions from 9:00 to 11:00 in the morning and 3:00 to 5:00, using two cameras (Canon EOS-1100D and Canon EOS-600D) with resolutions of 4272 × 2848 and 5184 × 3456. Sample images are shown in (Fig. 7). The second dataset contains images of each bunch taken from different angles before harvest, bunches on graph paper after harvesting, and videos of several palm and weight measurements such as height, weight, trunk circumference, total yield, and the number of bunches. This dataset has been developed for pre-harvest and harvest use and can use in many applications, including fruit detection, maturity analysis, Etc. In this experiment, to train the YOLO-V3 network, 2581 images used are divided into 80% for training and 20% for testing. All images are manually labeling by making a rectangular box covering the area containing the dates using the labelImg image annotation tool [28] with Yolo format, then the annotation information in each image is stored in a text file to be ready for use.
66
Y. Zarouit et al.
Fig. 6. Overview of the proposed method for date fruit detection
3.2 Network Parameterization and Training Details To detect date fruit, we used the standard Yolo-v3 algorithm with Darknet [26] is an opensource neural network framework written in CUDA and C, pre-trained on ImageNet dataset [19]. Then we adapted the weights of this model to our one-class date fruit detection problem, using our set of labeled images re-scaled to a spatial resolution of 416x416x3. The training achieved during 2000 iterations used the optimizer SGD BurnIn of Darknet, with a learning rate equal to 0.001, the batch size set to 64, then the momentum and decay were set to 0.9 and 0.0005, respectively. The model trained and tested using Google Colab or Collaboratory is a cloud service offered by Google (free) and based on Jupiter Notebook. This platform allows us to train machine-learning models in the cloud for free. Therefore, without needing to install anything on our computer except a browser, it gives us free Gpu but, with one disadvantage, it gives us just 12 h then it disconnects. In this experiment, a TESLA T4 GPU of about 15 GB is used to train our model. (Table 2) summarizes some essential training parameters used to train YOLO-v3 in our experiments.
Dates Detection System for Automatic Harvesting Using Deep Learning
67
Fig. 7. Sample images of the dataset showing dates in different conditions Table 2. Main training hyperparameters used to train yolov3 networks. YOLO-v3 Training iterations
2000
Batch size
64
Optimizer
SGD Burn-In
Learning rate
0.001
Momentum
0.9
Decay
0.0005
3.3 Result and Discussion To evaluate our model, we use different standard performance metrics related to the quality of detections produced by YOLO-v3 networks. The metrics for evaluation of the trained model are defined as follows: Recall =
TP TP + FN
(2)
68
Y. Zarouit et al.
Precision =
TP TP + FP
(3)
where TP is a true positive (correct detection), FN is a false negative (miss), and FP is a false positive (false detection). To show the comprehensive performance of the model, the F1 score was used as a trade-off between precision and recall, defined in Eq. (4). F1 =
2 ∗ Recall ∗ Precision Recall ∗ Precision
(4)
Another object detection metric used in this study to evaluate our model is the Average Precision (AP). It can show the global performance of the model under different confidence thresholds, it is defined as follows in Eq. (5). AP = (rn+1 + rn ) Pinterp (rn+1 ) n
with Pinterp(rn+1 ) = max p(˜r )
(5)
r˜ :˜r ≥rn+1
where p(˜r ) p(˜r ) is the measured precision at Recall r˜ The graph below shows the model performance during the training. The chart shows the average loss and the mean average precision versus the iterations throughout the model training. From (Fig. 8), we can see that as long as the iterations increase, the average losses decrease exponentially during the training time. The value of average losses indicates how the model performs based on the summation of errors made for each iteration. In our case, the final average loss of our model is 0.5. Therefore, the model achieves better accuracy of detection after training.
Dates Detection System for Automatic Harvesting Using Deep Learning
69
Fig. 8. Graph of losses and mean average precision VS iterations
(Fig. 9) presents four maturity stages of dates with different colors (green, yellow, brown), sizes, textures, and types. As shown, the model successfully identified all the dates bunches. Harvesting robots faces several challenges related to light changing in a natural environment that makes fruit detection more difficult. (Fig. 10) illustrates some qualitative results corresponding to some examples of dates under different daylight conditions of contrast and shadow.
70
Y. Zarouit et al.
Fig. 9. Dates detection results on the testing set (case: four maturity stages).
Fig. 10. Dates detection results on the testing set (the case of dates under daylight condition).
(Fig. 11) shows the results of date bunches detection in particular cases. Images on the top show the date fruit covered by bags. On the bottom left, dates are a blur, and on the bottom right, dates with a size smaller than 10% of the image.
Dates Detection System for Automatic Harvesting Using Deep Learning
71
Fig. 11. Dates detection results on the testing set (the cases of dates covered by bags, blur, and bunches size).
Based on the results presented before, the model was able to find all the date bunches in the images under different conditions of light even though the dates were covering by bags. On top of that, the system detects dates in only 0.028 s, which is extremely fast. (Table 3) summarizes the training performance of YOLO-v3 trained on the database [27]. Table 3. Yolo-v3 training performance. Training Simples
Average loss
Training time(h)
2065
0.5
7.7
Table 4 shows the performance of the model after training the model during 2000 iterations on the database [27]. The mean average precision is 95.74%. Due to the lack of use datasets similar to the one used in our experiment applied in date fruit detection. And to the best of our knowledge, there is no machine vision research for date fruit detection in an orchard environment. It’s impossible to make an exact comparison with the related works on this topic. For such a purpose, (Table 5) shows a comparison of the results of our model and other fruit detection models in an orchard used other techniques and object detection algorithms with a different dataset.
72
Y. Zarouit et al. Table 4. Yolov3 performance. YOLO-v3 Test Simples
516
mAP
95.74%
Precision
0.93
Recall
0.93
Average IoU
74.00%
F1-score
0.93
Test time(s)
0.028
Table 5. Comparison of results of the proposed method and others methods. mAP
time(s)
The proposed method
95.7% 0.028
Fruit detection and segmentation for apple harvesting using visual sensor in orchards [29]
82.7% 0.032
Deep fruit detection in orchards [30]
93.3% 0.13
Fruit detection, segmentation and 3D visualization of environments in apple 84.4% 0.03 orchards [31]
4 Conclusions A real-time machine vision framework for date fruit detection for harvesting robots in an orchard was proposed based on deep learning. The proposed solution consists mainly in a deep learning model to detect and locate date fruit bunches. Transfer Learning with Fine-Tuning is used to classify dates. A deep learning model was investigated, namely YOLO-v3 You Only Look Once, with pre-trained CNN models DARKNET35. To build a machine vision system, we used a rich image dataset of dates representing the challenges in the orchard environment. The proposed method achieves height accuracy, which is 95.7%, and a final average loss of 0.5 with classification times of 0.28 s. The presented method can successfully detect dates in the palms under different natural conditions, giving the robot valuable information as an output which is the first step for the robot to make harvesting decisions. Future work includes augmentation of images used in the training and consideration of other CNN models toward adapting the optimal detector; also propose a solution to analyze maturity and harvesting decision of dates. Moreover, to implement the proposed solution, it will be more interesting to build our dataset for having the training dataset and the test dataset coming from the same distribution.
References 1. Food and agriculture organization corporate. http://www.fao.org/faostat/en/#data/QC
Dates Detection System for Automatic Harvesting Using Deep Learning
73
2. Ministry of agriculture and maritime fisheries. http://www.agriculture.gov.ma/ 3. Kapach, K., Barnea, E., Mairon, R., Edan, Y., Shahar, O.B.: Computer vision for fruit harvesting robots state of the art and challenges ahead. Int. J. Comput. Vis. Robot. 3(1/2), 4 (2012). https://doi.org/10.1504/ijcvr.2012.046419 4. Bulanon, D.M., Kataoka, T., Okamoto, H., Hata, S.I.: Development of a real-time machine vision system for the apple harvesting robot. In: SICE 2004 Annual conference. IEEE, Vol 1, pp 595–598 (2004) 5. Bac, C.W., Hemming, J., van Tuijl, B.A.J., Barth, R., Wais, E., van Henten, E.J.: Performance evaluation of a harvesting robot for sweet pepper. J. Field Robot. 34(6), 1123–1139 (2017). https://doi.org/10.1002/rob.21709 6. Yasukawa, S., Li, B., Sonoda, T., Ishii, K.: Development of a tomato harvesting robot. Int. Conf. Artif. Life Robot. 22, 408–411 (2017). https://doi.org/10.5954/icarob.2017.os22-1 7. Xiong, Y., Ge, Y., Grimstad, L., From, P.J.: An autonomous strawberry-harvesting robot: design, development, integration, and field evaluation. J. Field Robot. 37(2), 202–224 (2019). https://doi.org/10.1002/rob.21889 8. Al-Janobi, A.A.: Application of co-occurrence matrix method in grading date fruits. In: Proceedings of ASAE Annual International Meeting, pp 3024–3098 (1998) 9. Al-Janobi, A.A.: Date inspection by color machine vision. J. King Saud Univ. 12(1), 69–79 (2000) 10. Mohana, S.H., Prabhakar, C.J.: A novel technique for grading of dates using shape and texture features (2015). arXiv preprint arXiv:1501.01090. https://doi.org/10.48550/arXiv.1501. 01090 11. Haidar, A., Haiwei, D., Mavridis, N.: Image-based date fruit classification. In: 2012 IV International Congress on Ultra Modern Telecommunications and Control Systems. IEEE, pp. 357–363 (2012). https://doi.org/10.1109/icumt.2012.6459693 12. Manickavasagan, A., Al-Mezeini, N.K., Al-Shekaili, H.N.: RGB color imaging technique for grading of dates. Sci. Hortic. 175, 87–94 (2014). https://doi.org/10.1016/j.scienta.2014. 06.003 13. Muhammad, G.: Date fruits classification using texture descriptors and shape-size features. Eng. Appl. Artif. Intell. 37, 361–367 (2015). https://doi.org/10.1016/j.engappai.2014.10.001 14. Aiadi, O., Kherfi, M.L.: A new method for automatic date fruit classification. Int. J. Comput. Vis. Robot. 7(6), 692 (2017). https://doi.org/10.1504/ijcvr.2017.10006506 15. Fadel M (2007) Date fruits classification using probabilistic neural networks. Agricultural Engineering International: CIGR Journal 16. Hossain, M.S., Muhammad, G., Amin, S.U.: Improving consumer satisfaction in smart cities using edge computing and caching: A case study of date fruits classification. Futur. Gener. Comput. Syst. 88, 333–341 (2018). https://doi.org/10.1016/j.future.2018.05.050 17. Altaheri, H., Alsulaiman, M., Muhammad, G.: Date fruit classification for robotic harvesting in a natural environment using deep learning. IEEE Access 7, 117115–117133 (2019). https:// doi.org/10.1109/access.2019.2936536 18. Mohamed, O., Khalid, E.A., Mohammed, O., Brahim, A.: Content-based image retrieval using convolutional neural networks. In: Mizera-Pietraszko, J., Pichappan, P., Mohamed, L. (eds.) Lecture Notes in Real-Time Intelligent Systems. RTIS 2017. Advances in Intelligent Systems and Computing, vol. 756, pp. 463–476. Springer, Cham (2019). https://doi.org/10. 1007/978-3-319-91337-7_41 19. Deng, J., Dong, W., Socher, R., Li, L-J., Kai, L., Li, F-F.: ImageNet: a large-scale hierarchical image database. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 248– 255 (2009). https://doi.org/10.1109/cvpr.2009.5206848 20. Girshick, R., Donahue, J., Darrell, T., Malik, J.: Rich feature hierarchies for accurate object detection and semantic segmentation. In: IEEE Conference on Computer Vision and Pattern Recognition, pp. 580–587 (2014). https://doi.org/10.1109/cvpr.2014.81
74
Y. Zarouit et al.
21. Lin T-Y., Goyal, P., Girshick, R., He, K., Dollar, P.: Focal loss for dense object detection. In: IEEE International Conference on Computer Vision (ICCV), pp. 2980–2988 (2017). https:// doi.org/10.1109/iccv.2017.324 22. Redmon, J., Divvala, S., Girshick, R., Farhadi, A.: You only look once: unified, real-time object detection. In: IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 779–788 (2016). https://doi.org/10.1109/cvpr.2016.91 23. Ren, S., He, K., Girshick, R., Sun, J.: Faster R-CNN: towards real-time object detection with region proposal networks. IEEE Trans. Pattern Anal. Mach. Intell. 39(6), 1137–1149 (2017). https://doi.org/10.1109/tpami.2016.2577031 24. Liu, W., et al.: SSD: single shot multibox detector. In: Leibe, B., Matas, J., Sebe, N., Welling, M. (eds.) Computer Vision – ECCV 2016. ECCV 2016. Lecture Notes in Computer Science, vol. 9905, pp. 21–37. Springer, Cham (2016). https://doi.org/10.1007/978-3-319-46448-0_2 25. Redmon J., Farhadi, A.: Yolov3: An incremental improvement (2018). arXiv preprint arXiv: 1804.02767 26. AlexeyAB/Darknet. https://github.com/AlexeyAB/darknet 27. Altaheri, H., Alsulaiman, M., Muhammad, G., Amin, S.U., Bencherif, M., Mekhtiche, M.: Date fruit dataset for intelligent harvesting. Data Brief 26, 104514 (2019). https://doi.org/10. 1016/j.dib.2019.104514 28. Open annotation tool LabelIm. http://labelme.csail.mit.edu/Release3.0 29. Kang, C.: Fruit Detection and segmentation for apple harvesting using visual sensor in orchards. Sensors 19(20), 4599 (2019). https://doi.org/10.3390/s19204599 30. Bargoti, S., Underwood, J.: Deep fruit detection in orchards. In: IEEE International Conference on Robotics and Automation (ICRA), pp. 3626–3633 (2017). https://doi.org/10.1109/ icra.2017.7989417 31. Kang, H., Chen, C.: Fruit detection, segmentation and 3D visualisation of environments in apple orchards. Comput. Electron. Agric. 171, 105302 (2020). https://doi.org/10.1016/j.com pag.2020.105302
Machine Learning for Diabetes Prediction A Systematic Review and a Conceptual Framework for Early Prediction Sara Retal1,2(B) , Hajar Sahbani1,3 , Nassim Kharmoum2,4,5 , Wajih Rhalem5,6 , and Mostafa Ezziyyani7 1
SmartiLab, Moroccan School of Engineering Sciences (EMSI), Rabat, Morocco 2 IPSS Team, Faculty of Sciences, Mohammed V University in Rabat, Rabat, Morocco [email protected] 3 LRIT, Faculty of Sciences, Mohammed V University in Rabat, Rabat, Morocco 4 National Center for Scientific and Technical Research (CNRST), Rabat, Morocco 5 Moroccan Society of Digital Health, Rabat, Morocco 6 E2SN Research Team, ENSAM, Mohammed V University in Rabat, Rabat, Morocco 7 Faculty of Sciences and Techniques of Tangier, Abdelmalek Essaadi University, Rabat, Morocco
Abstract. Diabetes is one of the most common diseases in the world for which there is still no cure. Every year it costs a lot of funds to treat people with diabetes. The prediction of diabetes will play a very significant role. This prediction has to be very accurate by using machine learning methods in order to maximize the accuracy of the prediction. However, the performance of the prediction applications relies on the suitable choice of the machine learning method. This paper aims to propose a systematic literature review of machine learning methods applied to diabetes prediction, revealing which are being studied in this field and the performance of the current state-of-the-art techniques. This paper provides the main results, challenges and opportunities, as well as a proposition of a conceptual framework for early prediction. Keywords: Machine learning intelligence
1
· Diabetes prediction · Artificial
Introduction
After a meal, the sugar contained in the food is partly transformed into glucose. This passes into the bloodstream, increasing blood sugar levels. To lower this blood sugar level, the pancreas produces a hormone: insulin. This hormone binds to a receptor present in the body’s cells and allows glucose to pass from the blood to the cells to be metabolized (used or stored). A blood glucose level is said to be normal if it is between 0.70 g/l and 1.10 g/l in the fasting state, or if it is lower than 1.40 g/l after a meal, and diabetes is diagnosed if the fasting blood glucose c The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 75–83, 2023. https://doi.org/10.1007/978-3-031-35248-5_7
76
S. Retal et al.
level is higher than 1.26 g/l on two occasions or if it is higher than 2 g/l at any time of the day Thus, diabetes is a chronic disease associated with abnormally high levels of glucose in the blood. It is caused by one of two mechanisms: – Insufficient production of insulin in the body. – Insufficient sensitivity of the cells to the action of insulin. Diabetes is ”one of the world’s leading killers”, along with high blood pressure and smoking, according to the World Health Organization (WHO). The disease is a major public health problem and despite prevention efforts, the pandemic continues. In 2014, diabetes affected 422 million people worldwide, while it affected only 108 million patients worldwide in 1980 and the first forecasts of the World Health Organization (WHO) and the International Diabetes Federation (IDF) worried in 1990 about the risk of diabetes affecting 240 million people in 2025. In 2019, diabetes affects more than 463 million people worldwide, including 59 million in Europe [1]. In 2021, diabetes will affect more than 537 million people worldwide (1 in 10), including 61 million in Europe [1]. For this, the prediction of this chronic disease is very crucial. This prediction must be very accurate. Hence, a reliable method must be used that uses artificial intelligence algorithms and, in particular, the use of machine learning in order to maximize the accuracy of the prediction to know if a person may be diabetic or not. However, an early warning system is convenient to help the medical staff. Machine learning is a branch that covers the statistical part of artificial intelligence, it trains computers to solve problems by examining thousands of examples, learning from them, and then using that experience to solve the same problem in new situations. The two main types of learning are Supervised learning, which provides the ability to predict the output variables (Y) for a data set from the input variables (x). It is usually performed in the context of classification and regression. Classification is a problem that arises when the output variable is a category, while regression is a problem that arises when the output variable is a real value. Popular examples of supervised machine learning algorithms include decision trees, K-nearest neighbor, logistic/linear regression. Unsupervised learning has only input data (X), its objective is to model the inherent structure from the input data. It includes two categories of algorithms: clustering and association algorithms. Clustering is the process of grouping data points based on their similarities, while the association is the process of discovering relationships between the attributes of these data points. Here are some unsupervised machine learning algorithms: K-means, Hierarchical clustering. The remainder of this article is organized as follows: the next section presents the adopted research methodology, as well as its application to diabetes prediction, Sect. 3 illustrates and describes the proposed conceptual framework for diabetes prediction. Finally, the conclusion and future directions of this work are highlighted in the last section.
Machine Learning for Diabetes Prediction
2
77
Research Methodology
This paper considers the process research represented in Fig. 1. This study aims to analyze the relations between diabetes prediction and machine learning. To do this, we have conducted a systematic literature review (SLR) to synthesize articles dealing with these two keywords. The most commonly used process to produce a systematic literature review consists of six actions. First, we define the research problem and set the respective questions and objectives. Then, we identify criteria and collect data. Finally, we synthesize data and elaborate a report.
Fig. 1. Research process.
2.1
Define the Research Problem
In the problem definition, the choice of keywords for building the search terms was based on phrases commonly found in the literature and words related to this review (i.e., machine learning methods applied to diabetes prediction). For the execution of the literature review, precise keywords were prepared and used for each chosen database (IEEE Xplore and Scopus), as follows: “diabetes prediction” AND (“machine learning” or “deep learning”). 2.2
Setting Research Questions and Objectives
First, our main goal in this section is to define the systematic literature review objective. Our research’s aim consists of evaluating the impact of machine earning methods on diabetes prediction, by studying and reviewing existing literature. SLR is a convenient technique for studying scientific research areas since it furnishes guidelines for achieving a process of research, from selection to synthesis of the obtained results. The diabetes prediction problem is widely studied; however, few papers have been conducted after the emergence of artificial intelligence. In this vein, we plan to respond by the present study to the following questions: 1. What is the impact of machine learning on diabetes prediction? 2. What are the machine learning methods that are being used? 3. How the machine learning methods are employed in prediction applications? 2.3
Criteria Identification
The considered criteria to obtain the literature review in this study are as follows:
78
S. Retal et al.
Inclusion Criteria – – – – – –
Journal articles, conference proceeding, and book chapters; Indexed in IEEE Explore and Scopus; In English language; Works about diabetes prediction; Describing the methodology and presenting the results.
Exclusion Criteria – – – –
Theses, meeting abstracts and interviews; Works not related to diabetes prediction and machine learning; Works that do not present any type of implementation; Works dated before the year 2015.
Keywords Combinations – – – –
Diabetes prediction AND machine learning; Diabetes AND machine learning;; Artificial intelligence AND diabetes; Artificial intelligence AND diabetes prediction.
2.4
Pertinent Literature Extraction
At this step, we acquire relevant articles resulting from the application of the criteria identification predefined in the last section. Table 1 shows a description of the selected papers. 2.5
Literature Synthesis and Discussion
A machine learning model using regression predicted diabetes with an area under the receiver operating characteristics of 0.80% and 0.78% in the discovery set and validation set, respectively, is proposed in [2]. The authors proposed a prediction model including environmental chemicals accurately predicted diabetes. The aim of the study in [9] is to design a model that can predict the probability of diabetes in patients with maximum accuracy. Then, three machine learning classification algorithms, namely: Decision Tree, SVM and Naive Bayes, are used in this experiment to detect diabetes at an early stage. school. The experiments are performed on the Pima Indians Diabetes Database (PIDD) database. The results obtained show that Naive Bayes outperforms with the highest accuracy of 76.30% compared to other algorithms. The authors in [14] made predictions of diabetes based on personal lifestyle indicators. At first, they performed a ChiSquare independence test before applying the CART (Classification and Regression Trees) method for data classification. Afterward, they used the method of Cross-Validation, and removed the bias in the result. As a result, the precision of
Machine Learning for Diabetes Prediction
79
Table 1. Relevent literature articles. Reference Year Title [2]
2022 Environmental chemical exposure dynamics and machine learning-based prediction of diabetes mellitus.
[3]
2022 Machine learning technique to prognosis diabetes disease: random forest classifier approach.
[4]
2021 Early prediction of gestational diabetes mellitus in the Chinese population via advanced machine learning.
[5]
2020 Diabetes prediction using artificial neural network.
[6]
2020 Classification and prediction of diabetes disease using machine learning paradigm.
[7]
2019 A robust voting approach for diabetes prediction using traditional machine learning techniques.
[8]
2019 Predicting long-term type 2 diabetes with support vector machine using oral glucose tolerance test.
[9]
2018 Prediction of diabetes using classification algorithms
[10]
2018 Machine learning based unified framework for diabetes prediction.
[11]
2018 Type 2 diabetes mellitus prediction model based on data mining.
[12]
2018 Analysis and prediction of diabetes mellitus using machine learning algorithm.
[13]
2017 Diabetes prediction using medical data.
[14]
2015 Prediction of diabetes based on personal lifestyle indicators.
[15]
2015 Prediction and diagnosis of diabetes mellitus-A machine learning approach
the CART method is equal to 75%. Blood pressure is identified as a significant factor in diabetes, along with other indicators such as roadside eating, sleeping late at night, family history of diabetes, or er edit e and the level of physical activity of a person per day. The authors in [10] compared the performances of the 6 techniques of machine learning classification: Artificial neural network, Support vector machine, Logistic regression, Decision tree, random forests, and Naive Bayes. They evaluated their performance using the 10-fold validation technique. Second, they proposed a framework for diabetes prediction, monitoring, and enforcement. The experimental results show that the highest classification accuracy is 74% and the highest F1 measure is 0.74. Moreover, the application is able to classify patients according to their level of diabetes by collecting realtime data from various health care departments, such as the center of medical diagnosis, hospital, health monitoring devices and sensors, etc.
80
S. Retal et al.
In [8], the authors propose a predictive model based on the support vector machine (SVM). Their approach exhibits an average accuracy of 96.80% and a sensitivity of 80.09% got on a holdout set. In [11], diabetes is predicted using significant attributes, and the relationship between the different attributes is also characterized. Various tools are used to determine meaningful attribute selection and for clustering, prediction, and exploration of association rules for diabetes. A significant selection of attributes was performed using the principal component analysis (PCA) method. Their results indicate a strong association of diabetes with body mass index and with glucose level, which was extracted using the Apriori method. Techniques of artificial neural networks, random forests, and K-means have been implemented for the prediction of diabetes. The Artificial Neural Network technique provided an accuracy of 75.7% and may be useful in aiding healthcare professionals in treatment decisions. The authors in [5] predict diabetes using an artificial neural network. They present the state of the art in the field of diabetes prediction. They identify the dice of existing techniques, namely: naive Bayes, decision trees, and support vector machines, and they propose solutions for them. The proposed model performs an analysis of publicly available data collected from diabetic patients listing the causative factors of diabetes, the age groups most affected, work style, and eating habits. The model uses artificial neural networks to detect diabetes and identify its type. Finally, prediction is useful in preventing other health conditions such as retinopathy, kidney disease, and cardiovascular disorders that may occur due to diabetes. The dataset used is that of PIDD. The highest accuracy of 85.09% proves the efficiency of the proposed work. The main objective of the study in [6] is to develop a system based on machine learning to predict diabetic patients. They used logistic regression to identify risk factors for diabetes based on p-value and odds ratio. They adopted four classifiers such as nave bays, decision trees, adaptive boost, and random forests to predict diabetic patients. Three kinds of partition protocols (K2, K5, and K10) also adopted and repeated these protocols into 20 tracks. The performances of these classifiers are evaluated using the precision and the Area under the curve (AUC). Using the NHANES dataset, the logistic regression model shows that 7 out of 14 factors such as age, education, BMI, systolic pressure, diastolic pressure, direct cholesterol, and total cholesterol are risk factors for diabetes. The overall accuracy of the system based on machine learning is 90.62%. The combination of the feature selection based on logistic regression and the classifier based on random forests gives an accuracy equal to 94.25% and 0.95 AUC (Precision Calibration Certificate) for the K10 protocol. The combination of logistic regression classifier and random forests works best. This combination will be very useful in predicting diabetic patients. The objective of the study in [7] is to find a model that can predict the probability of diabetes in patients with maximum accuracy. Here, a set of machine learning classification algorithms namely are used in this experiment to detect diabetes at an early stage on the PIDD dataset. The results obtained are the Knearest neighbors, adaptive boost, the decision tree, the random forests, the classification of the support vectors, the amplification of the gradient, the Logistic
Machine Learning for Diabetes Prediction
81
regression, the Multilayer Perceptron, naive Multinomial Bayes, naive Gaussian Bayes, and X-Gradient Boosting which provide an accuracy of 81.85%, 76.62%, 77.92%, 75.32%, 83.12%, 76.63%, 81.82%, 84.42%, 68.83%, 80.52%, and 80.52% respectively.
3
Conceptual Framework for Diabete Prediction
Fig. 2. Framework overview.
Our proposed framework (Fig. 2) aims to provide an intelligent process to manage diabetes prediction. The general idea of the framework is to create a system which makes it possible to predict diabetes, and to send an alert to the diabetic patient. Thus, we will have an early warning system. The Pima Indian Diabetes Database is provided by the National Institute of Diabetes and Digestive and Kidney Diseases [16]. Its objective is to predict by diagnosis whether or not a patient has diabetes, based on certain diagnostic measures included in the data set. In particular, all the patients in this dataset are women of at least 21 years of Indian origin. These data can be used as public data to start new predictions. Private data is added as the system works. The data set consists of several medical predictor variables and one target variable, the outcome. Predictive variables include: age, number of pregnancies for the woman, body mass index, plasma glucose concentration, diastolic blood pressure, skin fold thickness of the triceps, serum insulin level and genetic function of diabetes. We then have 768 observations and 9 variables of which 8 are numeric and a boolean variable. After collecting the data, it is necessary to prepare and clean the data through several
82
S. Retal et al.
steps, i.e., features extraction and selection. After the selection and the extraction, the next step is to divide the data set into two parts: a training set and a test set. The training set (i.e. training unit) is used to train the model, and the test set (i.e. testing unit) is used to test the model and assess accuracy. The prediction unit uses a real-time data and give a result using the testing unit. A user can be either an Administrator, a Patient or a Healthcare Professional. The front end offers to add and manage data, add and edit patient data, predict diabetes and send notification to diabetic patient.
4
Conclusion
This paper studied a systematic literature review, covering the major works of diabetes prediction using machine learning techniques. As a result, we have identified that each proposed technique addresses a specific approach, as consequence, there are difficulties to compare it to other techniques. In addition, it is possible to note that the accuracy, which varies between 74% and 96.80% is the only way the results. Since, after the application of machine learning, diabetes prediction becomes increasingly possible and promising. Based on this data, it becomes achievable to design a conceptual framework for early prediction, focusing on the main functions defined by prediction models. This paper presents certain limitations, which open the way to new research prospects. We achieved a systematic literature review based on a slightly reduced number of studied articles, therefore, the results obtained must be extended and reinforced in the next contributions, to be able to develop a decision-making tool, in all phases of diabetes prediction, by integrating suitable approaches and developing appropriate machine learning algorithms that co-occur with the goal of the early prediction.
References 1. International diabetes federation. idf diabetes atlas, 10th edn. brussels, Belgium (2021). https://www.diabetesatlas.org. Accessed 2 Feb 2022 2. Wei, H., et al.: Environmental chemical exposure dynamics and machine learningbased prediction of diabetes mellitus. Sci. Total Environ. 806, 150674 (2022) 3. Palimkar, P., Shaw, R.N., Ghosh, A.: Machine learning technique to prognosis diabetes disease: random forest classifier approach. In: Bianchini, M., Piuri, V., Das, S., Shaw, R.N. (eds.) Advanced Computing and Intelligent Technologies. LNNS, vol. 218, pp. 219–244. Springer, Singapore (2022). https://doi.org/10.1007/978981-16-2164-2 19 4. Wu, Y.-T., et al.: Early prediction of gestational diabetes mellitus in the Chinese population via advanced machine learning. J. Clin. Endocrinol. Metab. 106(3), e1191–e1205 (2021) 5. Nitesh, P., Geeta, R., Vijaypal, S.D., Ramesh, C.P.: Diabetes prediction using artificial neural network. In: Deep Learning Techniques for Biomedical and Health Informatics, pp. 327–339. Elsevier (2020)
Machine Learning for Diabetes Prediction
83
6. Maniruzzaman, Md., Rahman, M.J., Ahammed, B., Abedin, M.M.: Classification and prediction of diabetes disease using machine learning paradigm. Health Inf. Sci. Syst. 8(1), 1–14 (2020). https://doi.org/10.1007/s13755-019-0095-z 7. Mahabub, A.: A robust voting approach for diabetes prediction using traditional machine learning techniques. SN Appl. Sci. 1(12), 1–12 (2019). https://doi.org/10. 1007/s42452-019-1759-7 8. Hasan, T.A., et al.: Predicting long-term type 2 diabetes with support vector machine using oral glucose tolerance test. Plos one 14(12), e0219636 (2019) 9. Deepti, S., Dilip, S.S.: Prediction of diabetes using classification algorithms. Procedia Comput. Sci. 132, 1578–1585 (2018) 10. Mahmud, S.H., Hossin, M.A., Ahmed, M.R., Noori, S.R.H., Sarkar, M.N.I.: Machine learning based unified framework for diabetes prediction. In: Proceedings of the 2018 International Conference on Big Data Engineering and Technology, pp. 46–50 (2018) 11. Han, W., Yang, S., Huang, Z., He, J., Wang, X.: Type 2 diabetes mellitus prediction model based on data mining. Inf. Med. Unlocked 10, 100–107 (2018) 12. Alehegn, M., Joshi, R., Mulay, P.: Analysis and prediction of diabetes mellitus using machine learning algorithm. Int. J. Pure Appl. Math. 118(9), 871–878 (2018) 13. Singh, D.A.A.G., Leavline, E.J., Baig, B.S.: Diabetes prediction using medical data. J. Comput. Intell. Bioinf. 10(1), 1–8 (2017) 14. Anand, A., Shakti, D.: Prediction of diabetes based on personal lifestyle indicators. In: 2015 1st International Conference on Next Generation Computing Technologies (NGCT), pp. 673–676. IEEE (2015) 15. Vijayan, V.V., Anjali, C.: Prediction and diagnosis of diabetes mellitus-a machine learning approach. In: 2015 IEEE Recent Advances in Intelligent Computational Systems (RAICS), pp. 122–127. IEEE (2015) 16. Pima indians diabetes database dataset. https://www.kaggle.com/uciml/pimaindians-diabetes-database, Accessed 02 Feb 2022
Technology as an Answer to the Trust Crisis in Mental Health Services - Digitization Serving Mental Health Care Systems El Mehdi Ghalim(B) and Abdelmajid Elouadi National School of Applied Science of Kénitra, Advanced Systems Engineering Laboratory, University Ibn Tofail, BP 242 Av. de L’Université, Kénitra, Maroc {elmehdi.ghalim,abdelmajid.elouadi}@uit.ac.ma
Abstract. Nowadays, digitalization has become a potentiality with an impact that no sector can afford to ignore. Mental health care systems suffer, in a structural way, from a problem of accessibility and supply where digitalization can intervene, to make everyone benefit from these services, and thus circumvent the crisis of trust. In this perspective, digital innovations - such as affective computing, neural dialogue systems or even emotional intelligence - can play an effective trust agent role. We will focus in this article on the affirmation of this potentialities, sharing some legal concerns, and then doing a review of several examples of this ongoing integration. Keywords: Digital psychiatry · Trust crisis in mental health services · Digitization of mental health professions · Trust agent · Affective computing · Neural dialogue system · Emotional intelligence
1 Introduction According to the report provided in 2017, and co-authored by the 40 eminent scientists constituting the commission of the WPA (World Psychiatric Association), on the future of psychiatry “The digital psychiatry revolution has arrived” [1]. Digital psychiatry is identified as a major priority for the psychiatry field, public policy, and future research that could help impact billions of people. With its unconstrained geographic reach and potential to improve the scalability of interventions, digital psychiatry holds the potential for step change in service delivery and the development of new mental health treatments. In this article, we will seek to make a review of the literature, which highlights the potentialities and the limits of this digitalization.
2 Technology Serving Mental Health Care Systems 2.1 Revolutionary Potential In the scientific journal The Lancet, digital technology is described, in 2018, as offering the means to fill the global mental health treatment gap [2]. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 84–90, 2023. https://doi.org/10.1007/978-3-031-35248-5_8
Technology as an Answer to the Trust Crisis
85
Already in 2015, a study by the Institute for Healthcare Informatics informed us that 29% of the activity of the health applications sector was centered on mental health [3]. We also learn that, since 2015, more than 10,000 mobile phone applications for mental health were already available for download and use. While some apps are relatively light, like those that promote lifestyle change and wellness, others are more intensive, like apps that effectively turn smartphones into medical devices [4]. Health applications for mobile phones, were valued by some specialists, in the mobile technology market, at 28 billion dollars in 2018 and should reach, according to the same source, 102.35 billion dollars by 2023, according to Knowledge Sourcing Intelligence [5]. Finally, in 2019, a new article in the Lancet states that there is unanimous agreement among healthcare providers, medical associations, the healthcare industry and governments that automation using digital technology could improve the delivery and quality of psychiatric care and reduce costs” [6]. 2.2 Legal Concerns The report, previously cited [1], produced by the WPA commission also raised great concern about the privacy, transparency and confidentiality of digital mental health tools. This concern is common and legitimate given that certain groups of users, as shown in the letter to the English minister in 2018 [7], may have concerns about data security and confidentiality, with fears of discrimination if digitized personal mental health records were stolen, disclosed or sold. Precisely, according to Gooding [8], the digitization of mental health involves problems of legal and ethical limitations, even more than those already posed by digitalization of other medicine specialties. First, mental health treatment does not necessarily involve physical manipulation of the person by the professional. This difference in care delivery compared to general health allows mental health professionals to use forms of communication technology – especially in online counseling – more than in other medical fields. Second, digital mental health initiatives often involve higher data sensitivity compared to other forms of health delivery, and thus greater difficulty and uncertainty in applying legal standards. Personal mental health information is particularly sensitive under the current forensic framework in almost every country in the world. Already, as with general health, digitalization can help a person to access social services, state aid and essential health services (such as care, preventive intervention and well-being services), but unlike most areas of general health, personal mental health information can influence criminal proceedings (to attribute guilt, mitigate a sentence…), involve discrimination (for example in employment) and, perhaps the more importantly, be used to activate state-sanctioned coercion and effect hospital detention or forced treatment. In this sense, personal data on mental health differs – qualitatively – from data on general health. There remains, however, a point of agreement in the general debates, namely that the current legal and regulatory structures of digital mental health technologies are inadequate. Health and data privacy laws in most countries often omit requirements for digital mental health initiatives. Where requirements exist, they generally lack detailed regulations or remain siloed. Clearly, digital technologies can increase the vulnerability
86
E. M. Ghalim and A. Elouadi
of individuals and communities to health and safety risks, through several means among which we cite the lack of confidentiality, third party organizations exploiting information against the interests of users, infringements to reputation or even discrimination. As digital mental health technology becomes part of the global landscape of mental health services, Martinez-Martin reminds us [9] that the legal framework for meeting ethical obligations, related to mental health care, must be examined at the light of the paradigm shift in relations, which underlies these obligations, while considering the new outlook resulting from this digitalization. Mental health technology devices can be integrated as substitutes for mental health professionals or as additional components of the therapeutic relationship. In doing so, there is an impact on ingrained traditional norms, such as the traditional idea that we have of the therapeutic relationship supposed to be between therapist and patient, or even the belief that information about mental health is mainly generated and maintained in a doctor-patient relationship. In practice, technology in mental health may involve interactions with users outside of health domains, and data may be analyzed and shared in complex ways that the patient does not anticipate or easily understand. For these reasons, trust, data protection and confidentiality, accountability and transparency, pose fundamental ethical challenges in the use of digital mental health technology. The need for reliable oversight and mechanisms, given the impact of risks, is a key liability issue to ensure the safety and effectiveness of digital mental health devices. The ethical obligations of confidentiality, responsibility and competence are anchored in the relationship of trust. Some specific features of mental health apps and devices could have an impact on building trust in the therapeutic setting, for digital mental health care, which raises challenges for existing legal frameworks to meet ethical obligations of privacy, confidentiality and informed consent. For example, the gathering of large amounts of fine-grained personal information by some technologies introduces a level of monitoring, and therapeutic interaction, that could impact trust. When an app or device is inserted into, or replaces, aspects of a mental health care interaction, it raises questions about what it means to trust a device or who, or what, exactly is part of the relationship of trust. For technology devices or applications, reliability, features, and actions that allow technology developers, and these digital mental health tools, to be perceived as trustworthy are important. While trust is a willingness to take risks, trustworthiness encompasses qualities, such as integrity and competence, that allow one person to place trust in another. Delineation of responsibilities and monitoring are important in establishing reliability. If a mental health app provides informational resources, there should be confidence in the quality and reliability of the information. 2.3 Trust in Mental Health Technology In the same book [9], Martinez-Martin explains to us that trust, as a concept, requires reconsidering the distribution of obligations and responsibilities, being necessarily affected by this new landscape of digital mental health. She questions the possibility of having a therapeutic relationship with a device. Reliability also, as a concept, pushes us to redefine the mechanisms and relational needs necessary for the use of digital technological tools, accessible to the general public, to meet the mental health needs of
Technology as an Answer to the Trust Crisis
87
the population in an ethical manner. Digitization therefore offers promising avenues for meeting mental health care needs. In contrast, when it comes to direct-to-consumer mental health apps, there is still a great need for research into approaches that maximize trust and transparency, to define ethical obligations to be met that will protect users and facilitate integration into therapies. Trust is a complex and multidimensional concept, the dimensions of which vary depending on the context. Trust is cited as an important issue in the fields of health and information technology, but remains a difficult concept to define. Generally speaking, the concept of trust implies the willingness of a person in a situation of vulnerability to take a risk despite uncertainty, by trusting the intentions and skills of others. In health care, trust typically includes perceptions of provider competence, confidentiality and privacy, dignity, safety, and freedom from coercion or stigma. Trust is often described as fundamental in the therapeutic relationship. Trust in health care is often based on the quality of the interpersonal relationship between patient and provider, as well as whether there is a precedent for trusting a professional or an institution. Trust in digital technology involves guarantees of confidentiality, transparency, security and the perception of the competence of the supplier and the technology used. Thus, the components of trust in mental health technology and in care providers overlap when discussing key ethical obligations, particularly safety, competence, efficacy, and respect for privacy. There is a need to better understand how trust takes place in consumers’ use of mental health apps and tools, including awareness of potential risks. For example, it would be problematic if people avoid using certain potentially effective apps due to general distrust, or conversely, to trick people into trusting apps that are not safe or trustworthy. With chatbots and conversational agents, for example, it is important to ensure that they are designed to interact in a beneficial way for people with specific health problems, and above all that they cannot cause aggravation of mental disorders in users.
3 Practical Examples of Technology Use for Mental Health Digital mental health technology encompasses several technological solutions, used in the field of mental health, and refers to the use of digital technologies in initiatives to personalize mental health treatments, whether it is for individuals or groups. Some exploitations of technologies falling into this broad category are relatively old, such as online consulting, while others are newer. We find for example: • GPS systems that have been used to track psychiatric patients, in court, in several US jurisdictions [10] • An electronic check for visits, used to record the specific times when a mental health service, home visit, begins and ends [11] • In criminal law, there is a case where Machine Learning (the automated learning of an artificial intelligence) was used to predict the probability that a person would commit an offence, in particular by taking into account their mental health problems [12] • Psychiatric medications with built-in sensors that track ingestion and record timing compliance with prescribed treatment [13]
88
E. M. Ghalim and A. Elouadi
• Precision psychiatric medicine, developed using Big-Data, to treat individuals based on the variability of their genes, environment and lifestyle [14] • The virtual psychologist screens for stress, anxiety and depression during conversations using emotional intelligence modules. It is aware of each user’s stress level and mood and uses this ability to respond with empathy [15] • Digital phenotyping has been promoted by some clinicians to provide passive assessment of behavior, mood and cognition by applying Machine Learning to physiological and biometric data collected by smartphone [16] • An app designed to track users’ mood and cognition and then identify early signs of depression along with other diagnostic categories. Thus, the MIT Technology Review describes Mindstrong Health as “the smartphone app that can tell you’re depressed before you know it” [17] • Schools and universities in the United States have used machine learning to monitor student data and detect, and then provide support to, students who appear to be in mental health crisis [18] These are just a few examples that show how substantial the potential is, provided that the legal, societal and industrial challenges attached to it are met.
4 Discussion We tried to review the multiple angles of perception for the digitalization of mental health care systems. Therefore, this literature review makes it possible to identify the major obstacles and then to suggest appropriate recommendations. Thus, among the shortcomings noted, which continue to obstruct using digitalization for a better access to mental care, we can cite: • Limitations, and legal inadequacy, to ethically manage the effects of digitalization of the mental health care system. • Harm that can occur if certain, non-qualitative, applications are used, without the proper supervision, for people suffering from mental disorders. Regarding recommendations, which can be put forward by the author at this stage, the following can be retained: • Create a monitoring commission, made up of multidisciplinary professionals and researchers: industrialists, sociologists, technological experts, mental health care practitioners and managers of the system in place. This commission would carry out regular studies of the available applications of digitalization and industrialization for the local mental health care system and then decide what is beneficial and what is harmful. • Update, based on the reports of the monitoring commission, the legal framework with the help of legal experts, so that it is continuously relevant and modern. • Develop a master plan for the socio-technological reform of the mental health care system, based on the reports of the monitoring commission, and plan the appropriate
Technology as an Answer to the Trust Crisis
89
measures for an industrialization of the sector which takes into account local sociological specificities, takes advantage of technological advances and optimizes the delivery of mental care while guaranteeing its quality and accessibility. • An official virtual psychologist app, could help those who don’t yet dare to ask for help to circumvent the stigma, then refer them to competent professionals as soon as the need arises. • A disintermediation platform could facilitate the delivery of mental health care services, but above all restore trust and legitimacy in the practitioner-client relationship.
References 1. Bhugra, D., et al.: The WPA-lancet psychiatry commission on the future of psychiatry. Lancet Psychiatry 4(10), 775–818 (2017). https://doi.org/10.1016/S2215-0366(17)30333-4 2. Patel, V., et al.: The lancet commission on global mental health and sustainable development. Lancet 392(10157), 1553–1598 (2018). https://doi.org/10.1016/S0140-6736(18)31612-X 3. Lyle, J., Aitken, M.: Patient adoption of mHealth: use, evidence and remaining barriers to mainstream acceptance. IMS Institute for Healthcare Informatics (2015). https://www.iqvia. com/-/media/iqvia/pdfs/institute-reports/patient-adoption-of-mhealth.pdf 4. Nicholas, J., et al.: Mobile apps for bipolar disorder: a systematic review of features and content quality. J. Med. Internet Res. 17(8), e198 (2015). https://doi.org/10.2196/jmir.4581 5. Wood, L.: Mobile health (mHealth) app market - industry trends, opportunities and forecasts to 2023. Knowledge Sourcing Intelligence (2017). https://www.businesswire.com/news/ home/20171215005299/en/Mobile-Health-mHealth-App-Market-Industry-Trends-Opport unities-and-Forecasts-to-2023---Research-and-Markets 6. Bauer M, et al.: Automation to optimise physician treatment of individual patients: Examples in psychiatry. The Lancet Psychiatry. 6(4) (2019). https://doi.org/10.1016/S2215-036 6(19)30041-0 7. Gaebler, S., Toko, M., Jenkinson, S., Wortham, R.: Joint letter to minister hunt. Hunt – My Health Record: call to suspend my health record roll out (2018). 7 Aug 2018. Accessed 16 June 2020. http://being.org.au/2018/08/joint-letter-to-minister-hunt-my-health-records/ 8. Gooding, P.: Mapping the rise of digital mental health technologies: emerging issues for law and society. Int. J. Law Psych. 67, 101498 (2019). https://doi.org/10.1016/j.ijlp.2019.101498 9. Martin, N-M.: Trusting the bot: Addressing the ethical challenges of consumer digital mental health therapy. Dev. Neuroethics Bioethics. 3(3) (2020). https://doi.org/10.1016/bs.dnb.2020. 03.003 10. Miller, S.: The use of monitoring conditions (GPS tracking devices) re CMX [2014] QMHC 4. Psych. Psychol. Law. 22(3) (2015) https://doi.org/10.1080/13218719.2015.1032875 11. Olowu, A.: Delivering proof of care at the point of care. How electronic visit verification can benefit clinicians, home health workers and patients. Health Manage. Technol. 36(4) (2015). PMID: 26357772 12. Loomis, E-L.: 881 N.W.2d 749, 767. State v. Loomis. Supreme Court of Wisconsin (2016). https://www.courts.ca.gov/documents/BTB24-2L-3.pdf 13. Whitefield K (2017). Ostuka and Proteus digital health resubmit application to FDA for first digital medicine. Otsuka Pharmaceuticals. https://www.businesswire.com/news/home/201 70523005555/en/Otsuka-and-Proteus-Digital-Health%C2%AE-Resubmit-Application-toFDA-for-First-Digital-Medicine 14. Fernandes, B., et al.: The new field of ‘precision psychiatry. BMC Medicine. 15(1) (2017). https://doi.org/10.1186/s12916-017-0849-x
90
E. M. Ghalim and A. Elouadi
15. Winata, G., et al.: Nora the empathetic psychologist, EMOS technologies Inc. In: Proceedings of the Annual Conference of the international Speech Communication Association, INTERSPEECH, pp: 3437–3438 (2017). https://hdl.handle.net/1783.1/88698 16. Martin, N-M., et al.: Data-mining for health: staking out the ethical territory of digital phenotyping. Npj Dig. Med. 1(68) (2018). https://doi.org/10.1038/s41746-018-0075-8 17. Metz, R.: The smartphone app that can tell you are depressed before you know it yourself, MIT Technology Review (2018). https://www.technologyreview.com/2018/10/15/66443/thesmartphone-app-that-can-tell-youre-depressed-before-you-know-it-yourself/ 18. Castle, L.: How a UA professor is using data to identify potential dropouts. The Arizona Republic (2018). https://www.azcentral.com/story/news/local/arizona-education/2018/ 03/26/university-arizona-predict-dropouts-student-id-card-data/420348002/
Artificial Intelligence at the Service of Precision Medicine Wafae Abbaoui1(B) , Sara Retal1,2 , Nassim Kharmoum1,3,4 , and Soumia Ziti1 1
IPSS Team, Faculty of Sciences, Mohammed V University in Rabat, Rabat, Morocco wafae [email protected], [email protected] 2 SmartiLab, Moroccan School of Engineering Sciences (EMSI), Rabat, Morocco [email protected] 3 National Center for Scientific and Technical Research (CNRST), Rabat, Morocco 4 Moroccan Society of Digital Health, Rabat, Morocco
Abstract. Recently, the ability of artificial intelligence to transform data into information has affected the field of medicine. Indeed, artificial intelligence (AI) has been implemented in disease diagnosis and prognosis, treatment optimization and outcome prediction, drug development, and public health. This technological and especially medical advancement results from the ability of machine learning algorithms to process multidimensional data. This paper aims to study and evaluate recent literature dealing with the application of machine learning and deep learning algorithms in healthcare with a focus on precision medicine applications.
Keywords: Artificial intelligence precision medicine · health care
1
· machine learning · deep learning ·
Introduction
Precision medicine is a recent concept that has been in use since 2011 to describe medical treatments that are tailored to the individual characteristics of each patient [1]. Precision medicine is a fast-growing field that currently offers individualized medical treatments and disease prevention interventions. To enhance medical efficacy, this mostly entails customizing proactive and preventative therapy [2]. Among the promises of precision medicine are the detection or prediction of disease, accurate diagnosis, and optimization of treatment, while ensuring speed, accuracy, and cost-effectiveness (Fig. 1).
c The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 91–103, 2023. https://doi.org/10.1007/978-3-031-35248-5_9
92
W. Abbaoui et al.
Fig. 1. The key objectives of precision medicine.
The rapidly advancing fields of artificial intelligence, machine learning, and other technologies enable precision medicine to use biological and extrinsic data to improve clinical management decision-making for an individual in real-time throughout the disease. Unlike traditional clinical decision-making, precision medicine relies on highly detailed profiling of the patient’s genetic, morphological, and metabolic makeup [1]. In the context of these advances, artificial intelligence is the key technology that can develop precision medicine. Artificial intelligence is a branch of computer science that deals with machines that perform tasks requiring “human intelligence” [3]. To create an effective AI algorithm, computer systems initially receive data that is typically organized, indicating that each data point has a recognizable label or annotation by the algorithm [4]. Notably, The artificial intelligence we consider in this context includes machine learning and deep learning (Fig. 2).
Fig. 2. Artificial Intelligence vs Machine Learning vs Deep Learning.
Artificial Intelligence at the Service of Precision Medicine
93
Artificial intelligence technology has progressed rapidly in recent years and is being implemented in society. In the medical field, there is no exception [5]. In particular, AI is becoming increasingly significant in biomedicine and healthcare, where it has enabled improvements in biomedical image analysis, prognosis, patient care, clinical decision support, and AI is expected to play a crucial role in realizing the current global trend of precision medicine [6]. As an example, we can cite the effective contribution of artificial intelligence to the COVID-19 pandemic. We can mention an AI and NLP system named “COVID-19 Taxila” [7] that uses thousands of publications, clinical trials, and other relevant sources to allow users to access the dynamic landscape of scientific literature related to COVID-19, to search and analyze the data. It is worth mentioning that publications and data are automatically updated [8]. The remainder of this article is organized as follows: the next section presents the adopted research methodology, Sect. 3 describes the results of the literature review and illustrates the application of machine learning methods in the field of precision medicine. Finally, the conclusion of this work is highlighted in the last section.
2 2.1
Research Methodology Literature Review Planning Protocol
This paper considers the following planning protocol for the review: Research Questions Q1. How are AI methods being employed in precision medicine applications? Q2. What ML algorithms are most commonly used to process medical data and provide healthcare professionals with important information? Q3. What are the medical disciplines that are most involved and benefit from progress in precision medicine? Q4. What data types are used to apply AI methods? Databases for Literature Searching. This study was conducted on Scopus, the well-known scientific literature database. Inclusion Criteria. Journal articles, conference proceeding, book chapters and company reports; I1. I2. I3. I4.
Indexed in the Scopus database; Written in English; Explaining the methodology; Cross-sectional studies investigating changes over years.
94
W. Abbaoui et al.
Exclusion Criteria E1. Works not focused on the role of artificial intelligence in the evolution and progress of precision medicine. E2. Works that do not present any type of experimentation or comparison results, and make only propositions. E3. Works dated before the year 2015. Quality Criterion. QC1. Papers that focus on the synergy between AI and precision medicine and its impact on the health system. 2.2
Pertinent Literature Extraction
Publication Distribution Among Journals and Conferences. Of the 33 papers selected, 16 are reviews, 15 are articles, plus one conference paper and one note (Fig. 3).
Fig. 3. The number of documents by type [9].
Citation Analysis. One of the quality measures of a published paper is the number of times the article has been cited by other researchers. To perform citation analysis and maintain the quality of the selected article, the Scopus platform was chosen to determine the number of citations for the selected papers in the review. Table 1 shows the top ten papers in terms of the number of citations among all the selected papers in descending order. The conference paper entitled “Deep visual-semantic alignments for generating image descriptions” [10] outperformed the others, with the highest number of recorded citations (2313). The second most cited article, with 686 citations, was the journal article “Novel subgroups of adult-onset diabetes and their association with outcomes: a data-driven cluster
Artificial Intelligence at the Service of Precision Medicine
95
analysis of six variables” [11]. The third highest citation was from the journal article entitled “Deep Learning for Health Informatics” [12] with 682 citations. Based on the citation analysis results, the h-index of all the selected papers is 25. Table 1. Summary of the top 10 cited articles. Title
Publication Year Citations
Deep visual-semantic alignments for generating image descriptions [10] Novel subgroups of adult-onset diabetes and their association with outcomes: a data-driven cluster analysis of six variables [11] Deep Learning for Health Informatics [12] Deep Learning in Medical Imaging: General Overview [13] Artificial Intelligence in Precision Cardiovascular Medicine [14] Using recurrent neural network models for early detection of heart failure onset [15] Predicting cancer outcomes from histology and genomics using convolutional networks [16] Digital mammographic tumor classification using transfer learning from deep convolutional neural networks [17] DeepSurv: personalized treatment recommender system using a Cox proportional hazards deep neural network [18] Radiomics strategies for risk assessment of tumour failure in head-and-neck cancer [19]
2015 2018
2313 686
2017 2017 2017 2017
682 433 310 314
2018
283
2016
268
2018
250
2017
193
The figure below represents an overview of the citations for the selected papers.
Fig. 4. The number of citations per year [9].
96
W. Abbaoui et al.
Publication Distribution Along the Years. The figure below shows the number of papers that have been published from 2014 to 2021. These statistics clearly express the growing interest in artificial intelligence applications and various algorithms that can help open new doors for effective and personalized healthcare worldwide. The articles were published in the well-established and highly indexed database “Scopus”, starting with four articles published in 2014, and 512 articles between 2015 and 2018, and then 1646 articles between 2019 and 2021 ; The analysis results reveal that the trend to publish in this research area increased from 2016. This trend may be supported by the fact that machine learning, deep learning models, and AI applications are ushering in a new era of data-centric discovery in healthcare. This would reveal a promising area of established research, which explains the massive investment of many technology companies, such as IBM, Apple, and Google, in healthcare analytics to take precision medicine to the next level and increase the accuracy and prediction of patient outcomes (Figs. 4 and 5).
Fig. 5. The number of documents by year in the Scopus database [9].
3 3.1
Results of the Literature Review Machine Learning Methods Analysis
Machine learning methods, combined with large electronic health databases, could provide a customized approach to medicine through enhanced diagnosis and prediction of individual reactions to therapies [20]. Because machine learning is capable of handling huge volumes of data with complex interactions between variables, it is more likely to promote precision medicine than conventional statistical approaches [21]. The majority of machine learning approaches may be divided into two types: supervised and unsupervised. Supervised machine learning is based on a defined outcome [21]. Supervised methods are great, for classification and regression. Recent instances include the detection of a lung nodule from a chest X-ray; anticoagulant therapy risk
Artificial Intelligence at the Service of Precision Medicine
97
estimate models [22].Conversely, unsupervised learning algorithms, are used to find hidden patterns in training data that aren’t labeled. The most prevalent techniques for assembling data into previously confusing bundles are clustering approaches in unsupervised learning, such as K-means clustering [23]. Other algorithms, such as reinforcement learning, can be thought of as a combination of supervised and unsupervised learning that aims to improve the accuracy of trial and error [22]. 3.2
Random Forests
The tradeoff between accuracy and interpretability has always been a problem for machine learning applications. Ensemble learning approaches, such as random forests, which combine numerous classifiers into a highly accurate but less interpretable model, frequently outperform decision trees [24]. Random forest algorithms have been used in coronary computed tomography angiography, heart failure (HF) readmission, and HF risk and survival prediction models in precision cardiovascular medicine [14]. In a precision oncology study, researchers aimed to increase the predictive ability of their prediction models by combining radiomic variables with clinical data. The first step in their method is to employ logistic regression on rapid extraction of radiomic variables. Random forests are then utilized to combine the radiomic (continuous inputs) and clinical (categorical inputs) information into a single classifier. The results obtained showed that the combination of clinical variables with the optimal radiomic variables via random forests had a positive impact on the prediction and prognostic evaluation of locoregional recurrences (LR) and distant metastases (DM) [19]. 3.3
Support Vector Machines (SVM)
The support vector machine is a supervised learning algorithm for categorizing data into two or more groups. The term ‘support vector’ simply refers to the margin that the algorithm utilizes to determine whether or not data belongs in a given category. SVM’s strength is that it can handle large data sets with many variables or dimensions. SVM has been applied to several data sets due to its versatility, from identifying tissue and cell types based on genetic microarray expression data to categorizing breast mammograms as having microcalcifications or not [25]. Given its advantages of regularization and convex optimization, the SVM is the most popular classification algorithm and often exhibits the highest performance ranks for most classification problems [13]. In neuroradiology, distinguishing post-treatment alterations such as radiation necrosis and pseudo-progression from genuine tumor progression/recurrence is a typical difficulty. In this regard, there are just a few studies that use AI to distinguish post-treatment alterations from the central nervous system (CNS) tumor development.
98
W. Abbaoui et al.
In a study of 31 patients with glioma treated with surgery and chemotherapy, the SVM classifier has been trained to diagnose pseudo-progression vs recurrence. The classifier for pseudo-progression has a sensitivity and specificity of 89.91 and 93.72%, respectively, with an AUC of 0.94 [26]. In dementia prognostic research, SVMs were present in 30/37 selected studies, in 38 proposed models, and were by far the most used machine learning technique, according to the results of the systematic literature review (SLR). These numbers take into account the traditional SVM as well as variations. SVMs were used in all 30 research to determine whether or not mild cognitively impaired (MCI) patients would develop alzheimer’s disease (AD) [27]. 3.4
Decision Tree
Decision-tree-based algorithms are a particularly helpful class of algorithms for small to medium-sized datasets. Decision trees classify data by iteratively splitting it depending on the most informative biomarker, which is defined as the biomarker that delivers the greatest ‘information gain’ in statistical terms. When decision trees are joined into ensembles, such as a random forest, they become much more valuable. Boosting, a technique for assigning additional weight to samples that are difficult to categorize during training, can be used to train these ensembles. Due to their versatility, ease of training, and ability to handle correlated or unnecessary features without overfitting, tree-based models are particularly popular [28]. Artificial intelligence has also been used to quickly predict patients at the highest risk, allowing them to be prioritized and potentially reducing the mortality rate. In a study in Wuhan, China, researchers from Tongji Hospital, analyzed the electronic health records (EHRs) of 375 discharged patients to use clinical measures as features and trained a gradient-boosted decision tree model. The accuracy of the system was 93%. Its utilization would make it possible for physicians to promptly identify critical situations and take appropriate action. The model also discovered three crucial clinical features: lactic dehydrogenase, lymphocyte count, and high sensitivity C-reactive protein [8]. 3.5
K-means
A data-driven cluster analysis utilizing the k-means algorithm and hierarchical clustering was used to study patients with newly diagnosed diabetes (n=8980) from the Swedish All New Diabetics in Scania cohort. Clusters were created based on the 6 variables anti-glutamate decarboxylase antibody, age at diagnosis, body mass index (BMI), glycated hemoglobin, and Homeostatic Model 2 estimates of β-cell function and insulin resistance, which were then linked to data from prospective patient records on the incidence of problems and medication prescriptions. The Scania Diabetes Registry (n = 1466), all new diabetics in Uppsala (n = 844), and the Vaasa Diabetes Registry (n = 3485) were used to replicate the findings.
Artificial Intelligence at the Service of Precision Medicine
99
This study is the first step toward more accurate and clinically meaningful classification by combining information from the diagnosis with information from the health care system, and it represents an essential step toward precision medicine in diabetes [11]. In another study, researchers used k-means clustering to propose a new, easyto-use, cluster-based classification of bipolar disorder (BD) severity that could help clinicians in personalized medicine and shared decision-making processes. It is worth mentioning that the result obtained is 12 profilers from 5 life domains that classified patients into five clusters. The profilers were: Number of hospitalizations and suicide attempts, comorbid personality disorder, body mass index, metabolic syndrome, number of comorbid physical illnesses, cognitive functioning, being permanently disabled due to BD, overall functioning and leisure-time functioning, as well as the patient’s perception of functioning and mental health. The researchers obtained preliminary evidence for the construct validity of the classification: (1) all profilers performed well, increasing significantly with cluster severity, and (2) more severe clusters required more complex pharmacological treatment [29]. 3.6
Deep Learning (DL)
Deep learning is a subset of machine learning and a type of artificial neural network (ANN) that is modeled after the multilayered human cognitive system. Indeed, adding additional hidden layers to a neural network allows a deep architecture to express more complex hypotheses because the hidden layers can capture nonlinear relationships [30]. For that, DL is an AI technology that has the potential to perform automatic lesion detection, suggest differential diagnoses, and compose preliminary radiology reports, and it can even be used as a personalized treatment recommendation system and a useful framework for further medical research. In a study conducted by Katzman et al. [18], they introduced DeepSurv, a Cox proportional hazards deep neural network and state-of-the-art survival method to model the interactions between a patient’s covariates and treatment efficacy to provide personalized treatment recommendations. In another study, Futoma et al. [31] compared different models in their ability to predict hospital readmissions from a large EHR database. Furthermore, they applied deep learning methods to the five conditions that the Centers for Medicare and Medicaid (CMS) use to penalize hospitals. In addition, DL has paved the way for personalizing healthcare by providing unprecedented power and efficiency in mining large, unstructured multimodal information stored in hospitals, cloud providers, and research organizations. As data becomes available, DL systems can evolve and provide results where human interpretation is difficult. This can make disease diagnoses faster and smarter and reduce uncertainty in the decision-making process. Yet data integration across health informatics disciplines may be a limitation of deep learning in the face of supporting the future of precision medicine [12].
100
3.7
W. Abbaoui et al.
Artificial Neural Network (ANN)
Due to their powerful self-learning and complex biological information processing capabilities, ANN models have been widely applied to disease diagnosis, imaging analysis, and prognosis prediction. The aim of a retrospective study conducted by Que et al. [32] is to establish a neural network model capable of predicting the long-term survival of gastric cancer patients before surgery, in order to assess the status of the tumor before surgery. In the field of plastic surgery, plastic surgeons could use ANNs to predict postoperative complications after craniofacial surgery in the same way that ANNs have been used to predict recurrent cardiovascular disease [33]. Overall, deep learning applications are a collection of approaches that utilize advanced ANN architectures. These approaches are capable of modeling and learning predicted correlations in a wide variety of data types, and they have the potential to revolutionize the future of omics research and precision medicine applications [34]. 3.8
Convolutional Neural Network (CNN)
In a preliminary study, Huynh et al. [17] demonstrated that convolutional neural networks have the potential for computer-aided diagnosis (CADx) by learning features directly from image data instead of using analytically extracted features, Since it is difficult to train CNNs from scratch for medical images due to small sample sizes and variations in tumor presentations, transfer learning can be used to extract tumor information from medical images via CNNs initially pre-trained for non-medical tasks, minimizing the requirement for massive data sets. In another study concerning histology, Mobadersany et al. [16] presented a computational approach to learning patient outcomes from digital pathology images and illustrated how survival convolutional neural networks (SCNNs) can integrate information from both histological images and genomic biomarkers into a single unified framework to predict outcomes over time and show predictive accuracy that surpasses the current clinical paradigm for predicting overall survival of patients diagnosed with glioma. The results of this work highlight the emerging role of deep learning in precision medicine and suggest an increasing utility of computational histology analysis in future pathology practice. In the area of precision radiology, Oakden-Rayner et al. [35] presented proofof-concept experiments to demonstrate how routinely acquired cross-sectional computed tomography (CT) imaging can be used to predict patient longevity as an indicator of overall individual health and disease status using computational image analysis techniques, and they demonstrated that deep learning with CNNs can be usefully applied to radiomics research, for which a convolutional neural network was designed to predict all-cause mortality. Although the architecture of this network is based on current standard practices in the field, several modifications were necessary to account for the unique aspects of CT image data. Based
Artificial Intelligence at the Service of Precision Medicine
101
on this work, it is notable that computational image analysis applied to routinely collected medical images has considerable potential to improve precision medicine initiatives. 3.9
Recurrent Neural Network (RNN)
Recurrent neural networks are a type of ANN that focuses on temporal data. RNNs can significantly increase natural language processing, handwriting, and speech recognition ability. The aim of the study in [15] is to propose a new predictive model framework for Heart Failure (HF) diagnosis. Using a 12-month observation window, the area under the curve (AUC) of the RNN model was 0.777, which is significantly higher than the AUCs of supervised machine learning algorithms, namely: logistic regression, k-nearest neighbours (KNN), SVM, and multilayer perceptron (MLP). According to the study’s findings, RNN is the best approach for predicting the diagnosis of HF. In another study, Karpathy et al. [10] presented a multimodal RNN architecture that takes an image as input and provides a textual description of it. Their experiments revealed that the generated phrases surpass retrieval-based baselines and produce meaningful qualitative predictions.
4
Conclusion
To conclude, the move toward a data-driven healthcare system will have farreaching consequences for patients, clinicians, and civil society. Indeed, precision medicine is impossible to achieve in clinical practice without the help of advanced AI methods. However, AI will only be able to fulfill its mission if it remains a safe, effective, and proven aid to treating patients and improving healthcare. Yet, many technical and regulatory hurdles must be overcome for the AI-driven approach to precision medicine to become a reality.
References 1. Klonoff, D.C., Florez, J.C., German, M., Fleming, A.: The need for precision medicine to be applied to diabetes. J. Diabetes Sci. Technol. 14(6), 1122–1128 (2020) 2. Ho, D.S.W., Schierding, W., Wake, M., Saffery, R., O’Sullivan, J.: Machine learning SNP based prediction for precision medicine. Front. Genetics 10, 267 (2019) 3. Adir, O., et al.: Integrating artificial intelligence and nanotechnology for precision cancer medicine. Adv. Mater. 32(13), e1901989 (2020) 4. Subramanian, M., et al.: Precision medicine in the era of artificial intelligence: implications in chronic disease management. J. Transl. Med. 18(1), 472 (2020). https://doi.org/10.1186/s12967-020-02658-5 5. Hamamoto, R., et al.: Application of artificial intelligence technology in oncology: towards the establishment of precision medicine. Cancers 12(12), 1–32 (2020)
102
W. Abbaoui et al.
6. Boniolo, F., Dorigatti, E., Ohnmacht, A.J., Saur, D., Schubert, B., Menden, M.P.: Artificial intelligence in early drug discovery enabling precision medicine. Exp. Opin. Drug Discov. 16(9), 991–1007 (2021) 7. Covid-19 taxila. http://covid19.taxila.io/. Accessed 06 Mar 2022 8. Santus, E., et al.: Artificial intelligence-aided precision medicine for COVID-19: strategic areas of research and development. J. Med. Internet Res. 23(3), e22453 (2021) 9. Scopus web site. https://www.scopus.com/home.uri. Accessed 16 Mar 2022 10. Karpathy, A., Li, F.: Deep visual-semantic alignments for generating image descriptions. In: Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 07-12-June-2015, pp. 3128–3137 (2015) 11. Ahlqvist, E., et al.: Novel subgroups of adult-onset diabetes and their association with outcomes: a data-driven cluster analysis of six variables. Lancet Diabetes Endocrinol. 6(5), 361–369 (2018) 12. Ravi, D., et al.: Deep learning for health informatics. IEEE J. Biomed. Health Inform. 21(1), 4–21 (2017) 13. Lee, J., et al.: Deep learning in medical imaging: General overview. Korean J. Radiol. 18(4), 570–584 (2017) 14. Krittanawong, C., Zhang, H., Wang, Z., Aydar, M., Kitai, T.: Artificial intelligence in precision cardiovascular medicine. J. Am. Coll. Cardiol. 69(21), 2657–2664 (2017) 15. Choi, E., Schuetz, A., Stewart, W.F., Sun, J.: Using recurrent neural network models for early detection of heart failure onset. J. Am. Med. Inform. Assoc. 24(2), 361–370 (2017) 16. Mobadersany, P., et al.: Predicting cancer outcomes from histology and genomics using convolutional networks. Proceed. Nat. Acad. Sci. United States Am. 115(13), E2970–E2979 (2018) 17. Huynh, B.Q., Li, H., Giger, M.L.: Digital mammographic tumor classification using transfer learning from deep convolutional neural networks. J. Med. Imaging 3(3), 034501 (2016) 18. Katzman, J.L., Shaham, U., Cloninger, A., Bates, J., Jiang, T., Kluger, Y.: Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural network. BMC Med. Res. Methodol. 18(1), 24 (2018). https://doi.org/ 10.1186/s12874-018-0482-1 19. Valli`eres, M., et al.: Radiomics strategies for risk assessment of tumour failure in head-and-neck cancer. Scientific Rep. 7(1), 10117 (2017) 20. Wilkinson, J., et al.: Time to reality check the promises of machine learningpowered precision medicine. Lancet Digit. Health 2(12), e677–e680 (2020) 21. Shah, S.J.: Precision medicine for heart failure with preserved ejection fraction: an overview. J. Cardiovasc. Transl. Res. 10(3), 233–244 (2017) 22. Noorbakhsh-Sabet, N., Zand, R., Zhang, Y., Abedi, V.: Artificial intelligence transforms the future of health care. Am. J. Med. 132(7), 795–801 (2019) 23. Peng, J., Jury, E.C., D¨ onnes, P., Ciurtin, C.: Machine learning techniques for personalised medicine approaches in immune-mediated chronic inflammatory diseases: applications and challenges. Front. Pharmacol. 12, 720694 (2021) 24. Valdes, G., Luna, J.M., Eaton, E., Simone, C.B., Ungar, L.H., Solberg, T.D.: MediBoost: a patient stratification tool for interpretable decision making in the era of precision medicine. Scientific Rep. 6, 37854 (2016) 25. Handelman, G.S., Kok, H.K., Chandra, R.V., Razavi, A.H., Lee, M.J., Asadi, H.: eDoctor: machine learning and the future of medicine. J. Internal Med. 284(6), 603–619 (2018)
Artificial Intelligence at the Service of Precision Medicine
103
26. Sotoudeh, H., et al.: Artificial intelligence in the management of glioma: era of personalized medicine. Front. Oncol. 9, 768 (2019) 27. Dallora, A.L., Eivazzadeh, S., Mendes, E., Berglund, J., Anderberg, P.: Machine learning and microsimulation techniques on the prognosis of dementia: a systematic literature review. PLoS ONE 12(6), e0179804 (2017) 28. Ko, J., Baldassano, S.N., Loh, P., Kording, K., Litt, B., Issadore, D.: Machine learning to detect signatures of disease in liquid biopsies-a user’s guide. Lab on a Chip 18(3), 395–405 (2018) 29. de la Fuente-Tomas, L., et al.: Classification of patients with bipolar disorder using k-means clustering. PLoS ONE 14(1), 0210314 (2019) 30. Yu, Y., Li, M., Liu, L., Li, Y., Wang, J.: Clinical big data and deep learning: applications, challenges, and future outlooks. Big Data Mining Anal. 2(4), 288– 305 (2019) 31. Futoma, J., Morris, J., Lucas, J.: A comparison of models for predicting early hospital readmissions. J. Biomed. Inf. 56, 229–238 (2015) 32. Que, S., et al.: Application of preoperative artificial neural network based on blood biomarkers and clinicopathological parameters for predicting long-term survival of patients with gastric cancer. World J. Gastroenterol. 25(43), 6451–6464 (2019) 33. Kim, Y.J., Kelley, B.P., Nasser, J.S., Chung, K.C.: Implementing precision medicine and artificial intelligence in plastic surgery: concepts and future prospects. Plastic Reconstruct. Surg. - Global Open 7(3), e2113 (2019) 34. Grapov, D., Fahrmann, J., Wanichthanarak, K., Khoomrung, S.: Rise of deep learning for genomic, proteomic, and metabolomic data integration in precision medicine. OMICS J. Integrative Biol. 22(10), 630–636 (2018) 35. Oakden-Rayner, L., Carneiro, G., Bessen, T., Nascimento, J.C., Bradley, A.P., Palmer, L.J.: Precision radiology: predicting longevity using feature engineering and deep learning methods in a radiomics framework. Scientific Reports 7(1), 1648 (2017)
Case-Based Reasoning Approach, Integrating Deep Learning for Patient Diagnosis Combined X-Ray with Symptoms Moulay Youssef Ichahane(B) , Noureddine Assad, and Hassan ouahmane Laboratory of Information Technologies LTI, ENSA Chouaib Doukkali University, El Jadida, Morocco [email protected]
Abstract. Case based reasoning (CBR) is one of the most widely used reasoning approaches in expert knowledge-centered domains such as the medical sector, due to the risk that a false diagnosis may generate the requirement for models to be explainable. However, these models depend considerably on the user input (symptoms) and the input of the patient’s radiology image. Deep learning approaches based on convolutional neural networks have been proved in several papers to be relevant in imaging processing. This work proposes a hybrid framework of casebased reasoning and deep learning to be applied to a diagnostic support system. In the paper, we propose to couple the power of CNNs in radiology image analysis with the user-centered approach of case-based reasoning models. In addition, the framework on which we based is modular and adapts to a wide variety of tasks, data and uses, in fact, the system will receive as input the radiological image of the patient combined with the different symptoms, and will propose to the experts one or more predicted diagnoses with a probability score for each. The expert can validate or not a choice, his decisions will be taken in charge by the system to be added to the knowledge base in the case of validation or stored in another separate base in the case of non-validation. An implementation of the approach for diagnostic support is provided. Keywords: Case-Based-Reasoning · Deep Learning · Convolutional Neural Networks · Healthcare · Diagnosis
1 Introduction Case-based reasoning systems represent an approach with great potential for the diagnosis of diseases, such as Case-based reasoning was defined solves problems by retrieving similar, previously solved problems and reusing their solutions. Experiences are memorized, as cases in a case base. Inspired by memory-based human problem solving, the CBR approaches integrate commonly investigated retrieval techniques, k-nearestneighbor retrieval or simply k-NN [1], or inductive approaches [2], knowledge guided approaches [3], and validated retrieval [4]. As known, diagnostic reasoning includes cognitive activities like gathering information, recognition of patterns, solving problems and decision-making. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 104–115, 2023. https://doi.org/10.1007/978-3-031-35248-5_10
Case-Based Reasoning Approach, Integrating Deep Learning
105
This paper proposes a hybrid CBR-deep learning model to tackle the problem of diagnosis, the architecture of our proposal consists of the use of CBR model assisted by deep learning models to improve its performance. The motivation behind using CBR in the diagnosis of patients that it is easy to follow and understand the inference process they conduct, but also reduce the effect of the explosion of the knowledge base, caused by cases accumulation, Therefore, it is a blended solution between a knowledge-based system, where the knowledge must be elicited, and a deep learning model predictions, Inspired by the framework of [5], used for Radiology Report Recommendation, In other words, we have given priority to CBR, and deep learning will take over afterwards if there is no solution in the knowledge base. The implementation part we will give more details on our approach. The remainder of the paper is organized as follows. Section 2 provides an insight into diagnostics systems. Section 3 presents an overview CBR and deep learning systems especially neural networks, while the proposed hybrid CBR-deep learning model for diagnosis architecture and implementation of the model are explained. In Sect. 4 draws conclusions and future work.
2 Overview of Diagnostics Systems, Knowledge-Based and Non-knowledge-Based Reasoning in Healthcare In view of the increasing challenges faced by health systems, the considerable increase in the number of diseases, due to medical and technological advances but also to the growing number of patients, it has become vital and crucial to develop new generations of more efficient health systems, which will only be possible with the improvement of the diagnosis of patients diseases, which will allow to offer the appropriate care and medication. For this purpose, Health structures have found themselves obliged to follow the digital conjuncture by integrating hospital management systems into their business models processes to ensure good governance. After this process of digitalization, healthcare providers started to look for ways to improve the performance and quality of healthcare delivery, Beyond the improvements provided by statistical models and business intelligence tools, which still giving good monitoring and management dashboards; in parallel a promising branch has emerged, namely artificial intelligence, AI at the core of the different processes, from diagnosis to drug manufacturing, report writing, laboratory and others. In our paper we will focus on the so-called CDSSs (clinical decision support system) have been classified into two classes, the first is knowledge-based and the second is non-knowledge based as defined by [6]. In the state of the art, we have found a classification according to the type of active or passive delivery. Our work, will be oriented towards a clear improvement in the accuracy of system predictions applied in diagnosis process, The resulting system will be a combination of the two classes, in fact, in a case based reasoning architecture, we have integrated deep learning as proposed by [5]. In the literature, we find that by combining the rules in the knowledge base with patient data, the knowledge-based approach can be applied to the diagnosis of several different diseases, In the state-of-art, the knowledge-based approach generally allows the diagnosis of many different diseases, while the non-knowledge-based approach often focuses on
106
M. Y. Ichahane et al.
Fig. 1. Knowledge based vs non-knowledge bases systems by [6]
a narrow list of symptoms, such as symptoms for a single disease. In contrast, nonknowledge-based approach, which is founded on machine learning of old experiences with a determined number of characteristics, will be intrinsically significant if we focus on a given disease.
3 Introduction to CBR, Deep Learning Systems 3.1 Case Based Reasoning Our proposal is to implement a hybrid deep learning and CBR system to assist in the diagnosis of new cases. The proposed model automatically generates suggestions based on case-based reasoning. Contrary to some methods studied in the state of the art, our model involves the user in the learning process of the system, by validating the confirmed cases or not, this bidirectional relationship will allow the improvement of the system. The CBR model of [7] as been proposed a system similar to human reasoning, which consists in facing a new problem, the human is frequently calling to the previous experiences. CBR systems have proven to be efficiency in a wide range of application domains, [7] and [8]. Consider that the most similar solution to the previous problem, adapted if necessary to take into account the differences in the problem descriptions, is selected as the proposed solution to the target problem [9]. Therefore, this methodology When a new problem arises, a possible solution is obtained by retrieving similar cases from the case base and studying their registered solutions [10]. Solving a new case using CBR is considered as an execution cycle, four steps in the following order define each cycle: retrieve, reuse, preserve and revise, there are several types of model presentation, Fig. 1 illustrates an overview. • The first step is retrieve: the development of the case base by adding new confirmed cases is the main purpose of the search in the case based database, in addition to that in this step the system identifies similar cases to suggest solutions to assist the diagnosis.
Case-Based Reasoning Approach, Integrating Deep Learning
107
• The second step Reuse: In the majority of the cases, in the first stage allows to recover a solution that corresponds to the requirements of the new problem, but these solutions frequently require adaptations in order to be used on new contexts, it can integrate a layer of intelligence for an automatic adaptation as explained by [11]. • The third step Revise: Once the diagnosis is confirmed or not, the system generates a new case, in other words, whatever user choice, even if it does not validate the suggestion. Therefore, this new case cannot be added to the base of cases without receiving corrections of errors, in order to prevent suggesting to the next cases with errors, by consequence causing a wrong diagnosis. For this reason, a revision phase is fundamental. The improvement of the system depends on revising new cases, in other words, assures the utility and the necessity of the cases included. • Last step Retain: A great challenge of CBR is to have an optimal case base, which is to contain the maximum possible diagnosis with a minimum number of cases, besides that the system stores the unconfirmed suggestions separately, in order to be recalled if necessary in future uses, thereby guarantee an improvement of the model accuracy. CBR models are enhanced with additional information, in addition to updates based on new cases, at this step the updates of the modules are performed, such as a replacement of cases, switching from regular expressions to machine learning models, re-training of an existing model, periodically or upon attaining a particular number of cases.
Fig. 2. Typical CBR life cycle comprising four stages by [10]
In conclusion, in the literature, several approaches have been proposed to improve the DSS (Decision Support Systems), varying the techniques used and combining them in different ways in a Fig. 2 you can find one of its taxonomies, In other words, in the state of the art we find many contributions that aim to improve CBR, optimizing rules operations and reducing the knowledge database, but also by integrating machine learning (ML), deep learning (DL) and reinforcement learning (RL) techniques into the various CBR functional blocks(retrieve, reuse, revise, retain) (Fig. 3).
108
M. Y. Ichahane et al.
Fig. 3. Taxonomy of some knowledge based approch by [12]
As mentioned our work does not focus on the classical methods used in the optimization of CBR, an architectural or any other kind of optimization, [13] in his work has tried through a so-called semi-automatic adaptation of the existing case base based on a new case and the decision of the expert based on modifications of decision rules proposed automatically. A very relevant contribution if we remain specific to some use cases, but is discussed in terms of generalizability, in fact, according to the World Health Organization, in many cases the symptoms and diseases overlap and may be characteristic of several diseases. 3.2 Deep Learning In the late 1970s, Stanford was one of the first institutions to launch a program called SUMEX-AIM (Stanford University Medical EXperimental computer for Artificial Intelligence in Medicine) focused on applications of artificial intelligence research to biological and medical problems, Progress plateaued because the ingredients(GPUs performance) and AI algorithms did not yet exist. Today, three major factors have contributed to the emergence of AI in healthcare, the significant development of AI algorithms, supported by a digital transformation in the healthcare sector and finally the development of graphics processing units following development of microprocessors technologies. Deep learning models are machine-learning models that organize parameters into hierarchical layers. Features are multiplied and added together repeatedly, with the outputs from one layer of parameters being forwarded into the next layer before a prediction can be made. It is commonly admitted that, parameters are the set of numbers within a model that are used to define the model function (look like coefficients of the function) that are adjusted or “trained” during the machine learning process to accurately transform inputs into desired outputs; in other words, They are just numbers that are multiplied and added in various ways. The combination of parameters with a set of inputs via multiplication and addition is known as a linear combination. We use also activation functions, In deep learning terminology, we often use the term activation functions to refer to the nonlinear transformations we use, and we call the result of a linear combination followed by an activation function as the activations associated with the parameters involved, as illustrated in Fig. 4, among the most used activation functions we find Sigmoid, ReLU. The neurons of a neural network are, essentially, miniature logistic regression models. A layer of a neural network consists of a set of neurons that each take the same input.
Case-Based Reasoning Approach, Integrating Deep Learning
109
Fig. 4. Taxonomy of brain tumor using deep learning By [14]
Fig. 5. Sigmoid and ReLU functions
Fully connected layer (also called a dense layer or a linear layer) is a set of neurons that take the same input. The outputs are a linear combination of the inputs and the parameters of the layer at each execution of the neurons. Each layer then passes its outputs to the next layer. We call architecture of a neural network is how these layers are organized. Au explained in Fig. 5 a dense neural network algorithm. Convolutional layers are what make CNNs. CNNs are trained just like dense neural networks, albeit computing the gradient for gradient descent becomes a slightly more complex procedure. CNNs are immensely popular in subdomains of medicine such as radiology and nuclear medicine.
110
M. Y. Ichahane et al.
DNN algorithm We split our data into train, validation, and 5. The model does not learn from these test datasets samples because we do not execute the 2. We take a pass through our training dataset: optimization step during this phase. Our model will make a prediction based Without the optimization step, the model on the sample’s features cannot update its parameters, which in turn We will compute the loss between the prevents learning model’s prediction and the sample’s 6. The validation set is a measure of how the label. The loss is a numerical value model will do “in the real world.” We save a representing how far the prediction is version of the model if it gives us the best from the label. Low loss is good, and validation performance we have seen so far high loss is bad 7. The process is repeated multiple times, each The model will then update its time with different training configurations. parameters in a way that will reduce the This is known as hyperparameter tuning. loss it produces the next time it sees 8. This module focuses on the optimization that same sample. This is known as an step, which is comprised of three optimization components: step 1. Loss 3. Periodically, for example after we have 2. Gradient Descent taken a pass through our training dataset, we 3. Backpropagation can evaluate our model on a validation set 4. In this phase, we assess if the parameters the model has learned produce accurate predictions on data that it has not yet observed, in other words the validation set. 1.
Fig. 6. DNN algorithm
4 The Proposed Architecture for the Improvement of the Diagnostic Support System The case-based reasoning systems have many application areas in healthcare sector, which have provided solutions for diagnosis and treatment of diseases, based on experiences. Consequently, for complex medical diagnosis, if patients have a complex disease, more medical domains have to be used for this. For example, the Pulmonary diseases are amongst the most prevalent diseases and patients with a psychiatric disorder have a higher risk of developing chronic obstructive pulmonary disease (COPD), lung cancer and tuberculosis (TB) compared with the general population and also need complex algorithm for diagnosis. The CBR-based expert system used the k-nearest neighbor (kNN) algorithm to search k similar case that focusing on the Euclidean distance measure, unsupervised learning uses machine-learning algorithms to analyze and cluster unlabeled data sets. These algorithms discover hidden patterns in data, this category of learning models are used for three main tasks: clustering, association and dimensionality reduction. The overview of framework proposed for prediction method is shown in Fig. 6. The framework consists of 5 processes for prediction to difference between pneumonia and COVID19: deep learning (CNN) process, retrieval process, reusing process, prediction process and retain process respectively. This framework has two important phases: the learning phase, consisting of the use of deep learning algorithms to discover unknown patterns from medical images of patients, then combine them with other symptoms, the prediction phase, consists of the search for similar cases based on the K-NN algorithm.
Case-Based Reasoning Approach, Integrating Deep Learning
111
Fig. 7. Proposed framework phases
In case the system does not find any solution in the case base to find the closest case in terms of similarity. Then if the similar cases exceed the threshold will be proposed to the practitioner, otherwise the system will use deep learning to find a solution, and finally give user recommendations with scores as illustrated in Fig. 7
Fig. 8. DNN algorithm
112
M. Y. Ichahane et al.
5 Methodology The proposed model is in the process of implementation phase, this implementation is used to assess the performance of the proposed framework. Two different radiology datasets are used: Novel Corona Virus 2019 and Chest X-Ray Images (Pneumonia). For each studied dataset, two different scenarios are considered, We have exposed the system to the two datasets, the system is evaluated according to the percentage of classification of the two categories according to previews scenario, searching in the scalar base with a minimum of cases and then the neural network, in the scenario we propose to the experts the most accurate result for validation. However, two similar problems descriptions have similar solution descriptions, indeed, modeling similarity, generally, we have found three methods to handle similarity measures (metrics), the first method use functions to compare two cases: sim(case1,case2) = [0..1] (Hamming Distance of two cases), the second method based local similarity measure: similarity on feature level (for example: Sim = R - , 4 = = 10? | = 100 - 6 | Sim = 94%), the last method, global similarity measure (similarity on case or object level), In this category we find 2 types of methods: combines local similarity measures and takes care of different importance of attributes (weights). Nevertheless, efficient case retrieval is essential for large case bases, different approaches depending on case representation and size of the case base; it is also important organization of the case base, linear lists, only adapted for small case bases, index structures for large case bases. In the literature we found several techniques used in the diagnosis, simple technique, we observe first symptoms (e.g. fever) and values are measured (e.g. temperature degree = 40°), the goal is to find the cause (e.g. high body temperature) and treatment strategy (e.g. administering an analgesic), Case-Based Diagnosis technique, A case describes a diagnostic situation and contains description of the symptoms and the cause description of a treatment strategy, solving a diagnostic problem, start by make several observations about new patient, there is no obligation to know all the features, The new patient is a case without the solution part. Similarity is assessed for each feature and features can have different weights (importance), after similarity computation by weight average for each case, we select select most similar cases, then we go to the next step, Reuse of case by adapting solution, looking at how the difference between the new case and the selected case will affect the solution obtained. If the diagnosis is correct, we store a new case in the case based database (Retain) (Fig. 8). According to world health organization, symptoms of COVID19 as illustrated in Fig. 9, and pneumonia symptoms are presented in Fig. 10.
Case-Based Reasoning Approach, Integrating Deep Learning
113
fever cough tiredness loss of taste or smell sore throat headache aches and pains diarrhea a rash on skin, or discoloration of fingers or toes red or irritated eyes difficulty breathing or shortness of breath loss of speech or mobility, or confusion chest pain
Most common symptoms Less common symptoms
Serious symptoms
Fig. 9. COVID19 symptoms by WHO Pneumonia symptoms
Cough Shortness of breath Fever, sweating and shaking chills Fatigue Chest pain Nausea, vomiting or diarrhea Confusion, especially in older adults
Fig. 10. Pneumonia symptoms by WHO
6 Conclusion A case-based reasoning has been applied for diagnosis diseases such as diabetes, leukemia and lung, premenstrual syndrome, and breast cancer, thyroid and stroke disease [15]. In this paper, we have purposed the CBR framework to give as most precision for diagnosis of COVID19 and Pneumonia, in addition to this, both diseases are considered dangerous for the health of individuals, and require different treatment strategy. There are two processes, which differ from the original case-based framework (learning process and prediction process). In this paper, the model depend considerably on the user input (symptoms) and the input of the patient’s radiology image The CBR models improve with additional information, and updates based on new cases, then Deep learning approaches based on convolutional neural networks have been proved in several papers to be relevant in imaging processing can increase the system predictions. we can also think about a multi-agent system inspired by [16, 17] which used a system for the regulation and prioritization decisions are made through cooperation, communication, and coordination between different agents [18, 19]. Finally, the proposed model is in the process of implementation phase. To conclude, we can say that the state-of-the-art presented in this article clearly highlights the need for a more efficiency and precision. Research will focus more on the exploration of problems such as multi-datasets inputs with various deep learning approaches [20].
114
M. Y. Ichahane et al.
References 1. COVER, P. E. H. T.M.: Nearest neighbor pattern classfication, vol. I, pp. 1–28 (2012) 2. Jedwabny, M., Bisquert, P., Croitoru, M.: Probabilistic rule induction for transparent CBR under uncertainty. In: Bramer, M., Ellis, R. (eds.) SGAI-AI 2021. LNCS (LNAI), vol. 13101, pp. 117–130. Springer, Cham (2021). https://doi.org/10.1007/978-3-030-91100-3_9 3. Gao, L., Liu, C., Arefan, D., Panigrahy, A., Zuley, M.L., Wu, S.: Medical knowledge-guided deep learning for imbalanced medical image classification (2021). http://arxiv.org/abs/2111. 10620 4. Gonzalez-Ferrer, A., Seara, G., Cháfer, J., Mayol, J.: Generating big data sets from knowledgebased decision support systems to pursue value-based healthcare. Int. J. Interact. Multimed. Artif. Intell. 4(7), 42 (2018). https://doi.org/10.9781/ijimai.2017.03.006 5. Amador-Domínguez, E., Serrano, E., Manrique, D., Bajo, J.: A case-based reasoning model powered by deep learning for radiology report recommendation. Int. J. Interact. Multimed. Artif. Intell. 7(2), 15–26 (2021). https://doi.org/10.9781/ijimai.2021.08.011 6. Sutton, R.T., Pincock, D., Baumgart, D.C., Sadowski, D.C., Fedorak, R.N., Kroeker, K.I.: An overview of clinical decision support systems: benefits, risks, and strategies for success. npj Digit. ed. 3(1), 1 (2020). https://doi.org/10.1038/s41746-020-0221-y 7. Sànchez-marrè, M.: Principles of case-based reasoning. In: Cycle, pp. 1–13 (1994) 8. Aamodt, A., Plaza, E.: Case-based reasoning: foundational issues, method ological variations, and system approaches. Artif. Intell. Commun. 7(1), 39–59 (1996). https://ibug.doc.ic.ac.uk/ media/uploads/documents/courses/CBR-AamodtPlaza.pdf 9. De Mántaras, R., McSherry, D., Bridge, D.: Retrieval, reuse, and retention in CBR, Iiia. Csic. Es pp. 1–32, November 2016. http://www.iiia.csic.es/~mantaras/RRR_paper_KER.pdf 10. Hassan, B.A.R., Yusoff, Z.B.M., Othman, M.A.H., Bin, A.S.: Information is available at the end of the Chapter, https://doi.org/10.5772/55358, “We are IntechOpen , the world ’ s leading publisher of Open Access books Built by scientists , for scientists TOP 1 %,” Intech, p. 13, (2012). https://doi.org/10.1039/C7RA00172J%0A, https://www.intechopen. com/books/advanced-biometric-technologies/liveness-detection-in-biometrics%0A. https:// doi.org/10.1016/j.colsurfa.2011.12.014 11. Policastro, C.A., Carvalho, A.C.P.L.F., Delbem, A.C.B.: Automatic knowledge learning and case adaptation with a hybrid committee approach. J. Appl. Log. 4(1), 26–38 (2006). https:// doi.org/10.1016/j.jal.2004.12.002 12. El-Sappagh, S., Elmogy, M., Riad, A.M.: A fuzzy-ontology-oriented case-based reasoning framework for semantic diabetes diagnosis. Artif. Intell. Med. 65(3), 179–208 (2015). https:// doi.org/10.1016/j.artmed.2015.08.003 13. Pusztová, L., Babiˇc, F., Paraliˇc, J.: Semi-automatic adaptation of diagnostic rules in the casebased reasoning process. Appl. Sci. 11(1), 1–18 (2021). https://doi.org/10.3390/app11010292 14. Lee, B., Ellahi, W., Choi, J.Y.: Using deep CNN with data permutation scheme for classification of Alzheimer’s disease in structural magnetic resonance imaging (SMRI). IEICE Trans. Inf. Syst. E102D(7), 1384–1395 (2019). https://doi.org/10.1587/transinf.2018EDP7393 15. Chantamit-o-pas, P., Goyal, M.: A case based reasoning framework for prediction of stroke disease, no. Mi (2007) 16. Ikidid, A., El Fazziki, A., Sadgal, M.: A multi-agent framework for dynamic traffic management considering priority link. Int. J. Commun. Networks Inf. Secur. 13(2), 324–330 (2021). https://doi.org/10.54039/ijcnis.v13i2.4977 17. Ikidid, A., Abdelaziz, E.F., Sadgal, M.: Multi-agent and fuzzy inference-based framework for traffic light optimization. Int. J. Interact. Multimed. Artif. Intell. p. 1, (2021). (in Press). https://doi.org/10.9781/ijimai.2021.12.002
Case-Based Reasoning Approach, Integrating Deep Learning
115
18. Ikidid, A., El Fazziki, A., Sadgal, M.: A fuzzy logic supported multi-agent system for urban traffic and priority link control. J. Univers. Comput. Sci. 27(10), 1026–1045 (2021). https:// doi.org/10.3897/jucs.69750 19. Ikidid, A., Abdelaziz, E.F.: Multi-agent and fuzzy inference based framework for urban traffic simulation. In: Proceedings of 2019 4th International Conference on Systems of Collaboration Big Data, Internet Things Secur. SysCoBIoTS 2019, (2019). https://doi.org/10.1109/SysCoB IoTS48768.2019.9028016 20. Bianchi, R.A.C., López De Màntaras, R.: Case-based multiagent reinforcement learning: cases as heuristics for selection of actions. Front. Artif. Intell. Appl. 215, 355–360 (2010). https://doi.org/10.3233/978-1-60750-606-5-355
AGRI-PREDI Prediction System of Climate Change Based on Machine Learning for Precision Agriculture in Mediterranean Region Maroi Tsouli Fathi1(B) , Ramz Tsouli Fathi2 , Sarah Khrouch1 , Loubna Cherrat3 , and Mostafa Ezziyyani1 1 Mathematics and Applications Laboratory, Faculty of Sciences and Techniques of Tangier,
Abdelmalek Essaadi University, Tetouan, Morocco [email protected] 2 Materials, Environement and Sustainable Development Laboratory, Faculty of Sciences and Techniques of Tangier, Abdelmalek Essaadi University, Tetouan, Morocco 3 National School of Commerce and Management, Abdelmalek Essaadi University, Tetouan, Morocco
Abstract. Agriculture is one of the sectors that data mining learned before becoming popular. Today, it helps in making smart decisions on a variety of agricultural challenges such as eliminating difficult manual tasks and predicting crop yields based on climate change data. We present a new AGRI-PREDI solution that uses new intelligent agroclimatic functionalities. Data is collected from several national and international databases, and we apply smart new rules to create new trusted features. After that, we build a mathematical model that will be trained and adapted to different machine and deep learning models like CART (Decision Trees), SVM (Support Vector Machines), and KNN (K-Nearest Neighbors), as well as models of deep learning. Such as MLP (Multi-Layer Perceptron) and CNN (Convolutional Neural Networks). The study of this article has been applied to Mediterranean olive growing. The results of our solution demonstrate that the proposed new rules are effective for crop yield prediction. Deep learning has the highest level of accuracy, with values of 97.945% for the CNN model and 93.216% for the MLP model, respectively. Due to its high efficiency and accuracy when the data increases. However, CART shows good efficiency due to its logical tree structure. Keywords: Agriculture · Prediction · Climate change · Machine learning · Precision agriculture · Mediterranean region
1 Introduction Agriculture is undoubtedly one of the most climate-dependent human activities. However, the latter has tried to overcome this dependence and therefore often seeks to modify © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 116–127, 2023. https://doi.org/10.1007/978-3-031-35248-5_11
AGRI-PREDI Prediction System of Climate Change
117
the environment, for example by rinsing, by standardizing the topography, by increasing the size of the plot, by reducing the landscape heterogeneity by ensuring that the varieties high-yielding varieties bred for a familiar environment are always in optimum growing conditions. In this approach, which assumes that it is still possible to control crop growth conditions, climate change appears as a new factor [2]. In these modern landscapes characterized by agricultural intensification, innovative systems are needed to satisfy the overall long-term increase in global demand for food [3] and wood while addressing environmental concerns [4]. Current forecasts suggest that the Mediterranean region is likely to be strongly impacted by climate change [5]. Projections indicate that these changes are likely to negatively affect ecosystems and agricultural production throughout the Mediterranean basin. North Africa is expected to face significant challenges in terms of climate change [6]. Smart agriculture based on adaptation to climate change is a practice that will be very useful for Mediterranean countries such as Morocco, Spain and Tunisia, especially in desert and hilly areas where agricultural land and water are rare [12]. Our aim to study the effect of climate change on olive cultivation in the Mediterranean regions based on new techniques of data mining and data analysis to predict in advance the damage caused by the climate change on the quality of crop yields will allow a better decision.
2 Area of Application Your contribution may be prepared in LaTeX or Microsoft Word. Technical Instructions for working with Springer’s style files and templates are provided in separate documents which can be found here: (ftp://ftp.springer.de/pub/tex/latex/svproc/). We need all source files (LaTeX files with all the associated style files, special fonts and eps files, or Word or rtf files) and the final pdfs of all of the papers. References are to be supplied as Bbl files to avoid omission of data during conversion from Bib to Bbl. Our preferred bibliographic styles are MathPhySci and Basic (please see the references at the end of this document for examples of these two styles). The volume editors, usually the program chairs, will be your main points of contact for the preparation of the volume. 2.1 Geographical Application: The Mediterranean Region The Mediterranean climate is characterized by its temperate air, more rarely subtropical, characterizing the land areas bordering the Mediterranean basin. The main characteristics of a temperate climate are: a rhythm with four wellcontrasted seasons; a hot, dry summer; a marked winter, although mild (the monthly average is never below 0 °C); a sometimes very rainy spring and autumn, with a clear advantage in the fall [16]. The Mediterranean-type climate thus has as a particularity the coincidence between the annual pluviometric minimum and the thermal maximum, which the ombrothermal diagrams clearly show. To put it another way, it is the only climate whose hottest season is also the one that receives the least precipitation. Summer drought corresponds, according to the method proposed by H. Gaussen, to periods when the average precipitation is less than twice the average temperature (P < 2T) [17].
118
M. Tsouli Fathi et al.
2.2 Type of Crop in Application: Olives The production of olive oil in the Mediterranean basin can be considered as a science and as a mixture of creativity and innovation; it is also an art of living and a unique and ancestral heritage handed down from generation to generation. This historical interaction between Mediterranean cultures, the environment and the culture of the olive tree has resulted in a distinct cultural identity, and it is for this reason that we have chosen this culture for our study. The integration of new technologies in the olive oil sector and research towards increasing yields and improving the quality of olive oil have been priorities in the agricultural strategy of Morocco, the integration of data mining, and all very important for the agricultural sector in Morocco.
3 Building a New Logical Rule Based on Climate Change Data The base mathematical calculations of our model are based on data that we acquire from several national and international institutions. These data contain different and heterogeneous weather and climate features, so there is a need to have one base mathematical model that we will use to be fit and trained with machine and deep learning algorithms. To do so, we define some preliminary logical ideas to be mixed with Köopen and Gaussen models. We Build knowledge data from climate models as follow: 1. Study and classificate the type of climate 2. Define the appropriate climate model 3. Look for the climatic rules that characterize the development of a plant chosen with this type of model 4. Compare these climate rules with the previous climate and the current climate 5. Build a new knowledge base depending on its rules 6. Predict subsequently the quality of the yield of this crop based on climate rules and agroclimatic data After assembling several climatic indices from several climatic databases, it is time to filter these data according to the need for our model in our case the cultivation of the olive tree requires certain climatic rules to develop as well as possible the climatic classification de Köppen was developed based on the relationship between climate and vegetation. From this classification it is possible to define the climatic conditions for each region and then establish a relationship between the climate and the appropriate vegetation. This approach was established by Köppen based on The Köppen classification has been used in several research works concerning the phenomenon of climate change and calculates its evolution, the diagnoses made from the Köppen classification have given very correct results over the years and it is even used now for follow its evolution at deferring time scales. As discussed in the previous paragraph, features can have only 3 discrete values − 1, 0,1.
AGRI-PREDI Prediction System of Climate Change
119
Table 1. Our new data. Characteristics group
Climate factor indicator
values
Functionality in our dataset
1
Gaussen rules
Gaussen aridity Index
– 1, 1
G1
A cumulative precipitation of less than 10 mm for an average temperature of less than 10 °C
– 1.1
G2
A cumulative precipitation of less than 25 mm for an average temperature between 10 and 20 °C
– 1.1
G3
A cumulative precipitation of less than 50 mm for an average temperature of between 20 and 30 °C
– 1.1
G4
A cumulative precipitation of less than 75 mm for an average temperature above 30 °C
– 1.1
G5
Precipitation 0.9.
5 5.1
Simulation and Results Data and Simulation
We used the MIT-BIH Scalp EEG Database to simulate and evaluate the hierarchical algorithm on epilepsy EEG data. The data was first processed by reducing the frequency 256 Hz 1 Hz to match the description of seizures on the Physionet website in seconds and to compare with [6] that also used a 1-second time frame recording. The simulation was done using the machine learning and data visualisation libraries in Python (Pandas, Numpy, Matplotlib and Sklearn). Data is first loaded from the EDF file into the python script, where the 23 channels are combined by computing their mean, we found this method to produce the most consistent results due to the variation of the signals in different regions for different seizure iterations, particularly for different patients. The values are then transformed to be all positive in order to remove the horizontal symmetry. The Qulra and SeCA algorithms are then applied to the combined signal. 5.2
Results and Discussion
In this section, we present and discuss the obtained results. In order to do so, we will apply the algorithms summarized in the following on a data set with multiple patients’ entries: In the first part we execute model definition where we study old instances of seizures and extract their defining features. In this part, we use the Qulra algorithm to identify the recordings which contain a seizure. Recordings where a seizure is in place are characterized by an inclined regression
790
M. Abdelbaki et al.
Fig. 1. Comparison between recordings with and without an epileptic seizure
line whereas in a seizure-free graph, the line is horizontal (See Fig. 1). Then we apply the SeCA algorithm to separate the ictal cluster from the pre-ictal cluster (See Fig. 2). At this point, all that is left is to store the pre-ictal cluster
Fig. 2. Segmentation of data signals using the SeCA algorithm
values in a model to be used in the future as a basis for comparison with new signals. The second is where we use the extracted features to compare with newly collected data and look for similarities and indications of incoming seizures using the RT2CA algorithm which calculates a correlation ratio between the new data and the stored models. A high correlation ratio (CR > 0.9) means high likelihood of an incoming seizure. The proposed adaptation of the algorithm gives promising results since both of the outcomes of Qulra and SeCA are coherent with what we theorized.
A Hierarchical Machine Learning Algorithm for Epileptic Seizure Prediction
6
791
Conclusion
In this paper we explored the usability of the hierarchical algorithm for Junctional Tachycardia seizures’ detection proposed in [6] in the prediction of epileptic seizures. We theorized that the similarities in data structure and of the classification task could allow for an adaptation of the algorithm that was used to predict junctional tachycardia to predict epileptic seizures. The first results described in this paper prove the theory to be right. Thus, this discovery opens the door for multiple possibilities when it comes to the tasks of prediction using machine learning techniques in general, and for the specific task of predicting sudden emergencies related to the heart or the brain. Nevertheless, more research is yet to be done to confirm, on the end of extracting key features, the preliminary results both for different types of tasks and data sources and types. On the other end, correlation computing might not always be enough to conclude similarity of the signals, and so other methods are to be explored. We intend to conduct further research by using varying the data sources and classification tasks using the same algorithms to study their capabilities and limitations.
References 1. Moshe, S., Perucca, E., Ryvlin, P., Tomson, T.: Epilepsy: new advances. Lancet 385, 884–898 (2014). https://linkinghub.elsevier.com/retrieve/pii/S0140673614604566 2. Ikidid, A., et al.: A multi-agent framework for dynamic traffic management considering priority link. Int. J. Commun. Netw. Inf. Secur. 13(2), 324–330 (2021). https://doi.org/10.54039/ijcnis.v13i2.4977 3. Zhang, Z., Parhi, K.K.: Low-complexity seizure prediction from iEEG/sEEG using spectral power and ratios of spectral power. IEEE Trans. Biomed. Circuits Syst. 10(3), 693–706 (2016). https://doi.org/10.1109/tbcas.2015.2477264 4. Khan, H., Marcuse, L., Fields, M., Swann, K., Yener, B.: Focal onset seizure prediction using convolutional networks. IEEE Trans. Biomed. Eng. 65(9), 2109–2118 (2018). https://doi.org/10.1109/tbme.2017.2785401 5. Truong, N.D., et al.: Convolutional neural networks for seizure prediction using intracranial and scalp electroencephalogram. Neural Netw. 105, 104–111 (2018). https://doi.org/10.1016/j.neunet.2018.04.018 6. Affer, A., et al.: A convolutional gated recurrent neural network for epileptic seizure prediction. https://doi.org/10.1007/978-3-030-32785-9 8 7. Aitzaouiat, C.E., Latif, A., Benslimane, A., Chin, H.-H.: Machine learning based prediction and modeling in healthcare secured internet of things. Mob. Netw. Appl. 27(1), 84–95 (2021). https://doi.org/10.1007/s11036-020-01711-3
Epileptic Seizure Prediction Using Artificial Intelligence Methods Ilyas Zidane1(B) , Jamal Mhamdi1 , Mostafa Ezziyyani2 , Wajih Rhalem1,3 , and Nordine Zidane1 1 Mohammed V University, ENSAM, Rabat, Morocco
{j.elmhamdi,w.rhalem,n.zidane}@um5r.ac.ma 2 Abdelmalek Essaadi University, FST, Tangier, Morocco [email protected] 3 Moroccan Society of Digital Health, Casablanca, Morocco
Abstract. Epileptic seizures are caused by sudden electric discharges in neurons irregular transient operation; hence this generates cerebral neurological dysfunction. The seizures can be mortal because it occurs unexpectedly. Artificial intelligence prediction methods could avoid any injuries relatives to the seizure. There are several causes of seizures. It can be linked to a disease (a genetic disease, a mental disorder, a neurological infection, etc.) or occur after an external event (falls, drug addiction, chronic alcoholism, etc.). Some people are born with epilepsy, but some may have epilepsy for the rest of their lives. A person can have epilepsy seizure, even though he is not epileptic. An epileptic seizure is not fatal but death can happen accidentally, in the event of a fall or drowning for example. Thus, why a prediction of the time of epilepsy seizure is necessary. Patient’s age and Gender are important, the main cause for adult patients is stress and particularly for women patients is hormonal disorder, however the psychic side of the patient is likely to cause an epileptic seizure. In this paper, we propose a new approach of epilepsy seizure prediction by using Artificial Intelligence methods to analyze seizures of a huge number of patients EEGs. This method is most powerful and most accurate than the others prediction methods. Keywords: Epilepsy · Seizure · EEG · Artificial Intelligence · Supervised learning · Algorithms
1 Introduction The statistics shows nearly 5% of persons are experienced a seizure in their life time and 1% of persons suffer of epileptic seizure [1]. In this paper we will apply the methods of artificial intelligence on electroencephalograms of 150 epileptics patients of all categories of age: Children (5–14 years), Youth (15–24 years), Adults (25–64 years) and Seniors (65 years and over). Patients mostly worry about when and where the seizure happens. Many methods have been developed for seizure prediction over several years since 1995. The most of these methods used © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 792–804, 2023. https://doi.org/10.1007/978-3-031-35248-5_70
Epileptic Seizure Prediction
793
state similarity features for seizure prediction[2],also they employed independent component analysis to separate isolated epileptiform discharges from EEG background and analyse magnetoencephalography data to localize epileptic spikes using independent components analysis and spatio-temporal clustering [3],they used EEG spatiotemporal correlation structure with delay correlation and covariance matrices [4], they applied autoregressive modeling and least-squares parameter estimator [5], they studied effects of linear univariate features [6], Classification of patterns of EEG synchronization with bivariate features [7], they used spectral power from raw and bipolar time-differential signals [8] and combination of mean phase coherence and the dynamic similarity index methods [9, 10] , they used cost-sensitive support vector machines in spike rate [11] and they employed Fourier transformation to perform signal processing of epileptic EEG [12].
2 Data Set Description The data set is issued from EEGs recordings of epileptic’s patients. The information in large clinical EEGs database has been entered during three years because of the huge number of patients and their time’s unavailability. We choose only 150 patients to study the prediction of seizure using artificial intelligence methods so to predict life threatening caused by epileptic seizure and may be able to analyze signals in real-time and having a dashboard to generate alerts before the epileptic seizure happening. The sampling frequency is 256 Hz with 128 channels of epileptic EEG signals which are recorded by focal electrodes form brain locations: temporal lobe epilepsy (TLE), frontal lobe epilepsy (FLE), posterior epilepsy (PE including the parietal, occipital, occipito-temporal, temporo-occipito-parietal junction regions), operculo-insular epilepsy (OIE) and multifocal epilepsy (MFE).The records may be taken before the seizure (Preictal ), during the seizure(ictal), between seizures (interictal ) and after the seizure(postictal). Preictal or prodromeis the time before the seizure. It can last from minutes to days and make patient act and feel differently, for example, duration of ictal varies from a few seconds to 5 minutes. Therefore, ictal records during epileptic seizures can have at least 60 minutes of preictal signal preceding each seizure (Table 1). We have 150 persons with 240 EEG’s data points S1 to S240. The person may be male or female (Senior, Adult, Youth or child). We record the EEG signal of the brain when the patient had their eyes open in both cases no seizure and seizure activity. For each patient we put the number of seizures in dataset field named seizure and 0 if there are no seizures. Dataset represents of EEG signals belong to 50 healthy persons and 100 epileptic patients. EEG signals are separated into 5 frequency bands that are Delta (0–4 Hz), Theta (4–8 Hz), Alpha (8–16 Hz), Beta (16–32 Hz) and Gamma (32–64 Hz) by Wavelet Transform. Then the power spectrum from each band is computed to be the feature. Figures 1, 2 and 3 show general information about the patients who were selected as part of the dataset. The number of trials for the non-seizure class is: 50 The number of trials for the seizure class is: 100
794
I. Zidane et al. Table 1. Global dataset
Patient
Age
Gender
Seizures
S1
….
S240
1
Adult
Female
4
180
73
2
Adult
Female
0
−116
−30
3
Adult
Male
0
−3
−48
4
Adult
Female
5
−565
−99
5
Adult
Female
2
23
−157
6
Senior
Male
0
−37
34
7
Adults
Female
1
419
190
8
Young
Female
1
−30
35
149
Adult
Female
2
219
78
150
Adult
Male
1
−87
311
Fig. 1. Epilepsy seizure count.
The number of trials for the female class is: 87 The number of trials for the male class is: 63 The number of trials for the senior class is: 26 The number of trials for the adult class is: 75 The number of trials for the young class is: 26 The number of trials for the child class is: 23 In this paper our approach was decomposition into frequency bands by using Fourier Transform to the entire signal by means of Fast Fourier Transform with transformation from time domain to frequency domain to extract of relevant events (patterns, peaks, valleys, among others). Thus, we can identify epileptic and non-epileptic patients. Before using dataset for further processing, we must apply data preprocessing method to convert raw data into an understandable format in order to solve the problems in actual
Epileptic Seizure Prediction
795
Fig. 2. Female and male patient count.
Fig. 3. Age patient count.
data such as incomplete, inconsistent, missing or erroneous data. Preprocessors executes tasks in the following order: • • • • •
Removes instances with unknown target values Continues categorical variables (with one-hot-encoding) Removes empty columns Imputes missing values with mean values Normalizes the data by centering to mean and scaling to standard deviation of 1
796
I. Zidane et al.
We use exploratory data analysis (EDA) an approach to analyzing dataset statistics, it summarizes the pertinent characteristics. EDA show dataset beyond the formal modeling or hypothesis testing task. In orange software a method to use preprocessors is to compare the results of deep learning model with the default model Constant, which always predicts the majority class (Table 2). Table 2. Exploratory data analysis Patient count
S1
150,00
S2
S3
…
S239
S240
150,00
150,00
150,00
…
150,00
150,00
mean
75,50
15,40
−1,10
−32,44
…
−15,95
−30,77
std
43,45
269,84
249,02
269,52
…
348,29
233,70
min
1,00
−961,00
−812,00
−1021,00
…
−1850,00
−1269,00
25%
38,25
−80,00
−117,00
−112,50
…
−206,00
−99,00
50%
75,50
10,50
−5,50
−17,50
…
−21,50
−4,00
75%
112,75
129,50
71,75
53,75
…
44,00
74,50
max
150,00
1036,00
1235,00
1198,00
…
697,00
690,00
As we can observe Epileptic seizers has data values higher than non-Epileptic. Table 3. Min and max eeg’s values (4 patients) Patient
Min
Max
1
−353
242
YES
2
−130
70
NO
3
−91
58
4
−624
1638
Epilepsy state
NO YES
We use MATLAB software to plot 4 EEGs Fig.4 as a sample. EEG values in vertical axis and Data points S1-S240 in horizontal axis. Usually records of Epileptic seizers are smoother and looks like have a tendency (Table 3).
Epileptic Seizure Prediction
797
Fig. 4. EEGs of top 4 patients.
Figure 5 show comparison between the signal of an epileptic patient and a nonepileptic patient. Critical differences in the two signals cannot be distinguished given the nature of the signal.
Fig. 5. EEG patients 1 and 2
3 Artificial Intelligence Methods In this paper we use python artificial intelligence algorithms by using Anaconda and Orange Applications.
798
I. Zidane et al.
3.1 K-Nearest Neighbors Predict according to the nearest training instances (Table 4). Table 4. Inputs/Outputsn in KNN Inputs
Outputs
Data: input dataset
Learner: kNN learning algorithm
Preprocessor: preprocessing method(s)
Model: trained mode
The kNN algorithm searches for k closest training examples in feature space and uses their average as prediction. We set the number of nearest neighbors to 5, the distance parameter (metric) to Euclidean (“straight line”, distance between two points) it could be : Manhattan (sum of absolute differences of all attributes) or Maximal (greatest of absolute differences between attributes) or Mahalanobis (distance between point and distribution) and for weights we use Uniform: all points in each neighborhood are weighted equally it could be Distance: closer neighbors of a query point have a greater influence than the neighbors further away. The Fig. 6 is a regression task workflow shows how to use the Learner output. We used the Epilepsy dataset. We input the kNN prediction model into Predictions and observe the predicted values. In orange software we use preprocessors by comparing the results of KNN with the default model Constant, which always predicts the majority class.
Fig. 6. KNN Orange preprocessors Workflow
Epileptic Seizure Prediction
799
Table 5 shows probabilities that the score for the model in the row is higher than that of the model in the column. Compare models by Classification accuracy Repeat train/test: 10 Training set size: 66 %. Table 5. Inputs/Outputs in KNN KNN KNN
Constant 0.266
Constant
0.734
Small numbers show the probability that the difference is negligible. The confusion matrix Table 5 gives us counts of each of the types of results. It’s a table showing proportion of predicted (Table 6). Table 6. KNN Confusion Matrix Predicted Actual
NO
YES
S
NO
55.6%
0.0%
YES
41.1%
100.0%
100
S
90
60
150
50
Table 7 shows the KNN performance scores. AUC means area under the curve. Accuracy measures how many observations, both yes and no, were correctly classified. F1 score is the harmonic mean of precision and recall. Precision is a metric which determines the proportion of correctly classified yes identifications. Recall is defined as the proportion of yes that were correctly identified. Log Loss is the negative average of the log of corrected predicted probabilities for each instance (Fig. 7). Specificity is the proportion of yes that are correctly predicted by the model.
800
I. Zidane et al. Table 7. KNN performance scores
Model
AUC
CA
F1
Precision
Recall
LogLoss
Specificity
KNN
1.000
0.733
0.738
0.852
0.733
0.422
0.867
Fig. 7. KNN Orange predictions Workflow
3.2 Logistic Regression Logistic Regression learns a Logistic Regression model from the data. It only works for classification tasks Logistic regression, or logit regression, or logit model is a regression model where the dependent variable is categorical. Our paper uses the case of a binary dependent variable, thus take only two values, “0” and “1”, which represent outcomes yes/no (epileptic or not). Set model parameter: Regularization type (either L1 or L2). Set the cost strength (default is C = 1) 3.3 Support Vector Machine (SVM) Support vector machine (SVM) is a machine learning technique that separates the attribute space with a hyperplane, thus maximizing the margin between the instances of different classes or class values. The technique often yields supreme predictive performance results. Orange embeds a popular implementation of SVM from the LIBSVM package. For regression tasks, SVM performs linear regression in a high dimension feature space using an ε-insensitive loss. Its estimation accuracy depends on a good setting of C, ε and kernel parameters. The outputs class predictions based on a SVM Regression. Support Vector Machine (SVM) model is a Supervised Learning model used for classification and regression analysis. It is a representation of the examples as points in space, mapped so that the examples of the separate categories are divided by a clear gap that is as wide as possible. New examples are then mapped into that same space and predicted to belong to a category based on which side of the gap they fall.
Epileptic Seizure Prediction
801
In addition to performing linear classification, SVMs can efficiently perform a nonlinear classification using what is called the kernel trick, implicitly mapping their inputs into high-dimensional feature spaces. Suppose some given data points each belong to one of two classes, and the goal is to decide which class a new data point will be in. In the case of support vector machines, a data point is viewed as a p -dimensional vector (a list of p numbers), and we want to know whether we can separate such points with a (p−1) -dimensional hyperplane. When data are not labeled, supervised learning is not possible, and an unsupervised learning approach is required, which attempts to find natural clustering of the data to groups, and then map new data to these formed groups. The clustering algorithm which provides an improvement to the support vector machines is called support vector clustering and is often used in industrial applications either when data are not labeled or when only some data are labeled as a preprocessing for a classification pass. Model parameters • SVM type with test error settings. SVM and ν-SVM are based on different minimization of the error function. On the right side, you can set test error bounds: • SVM • Cost: penalty term for loss and applies for classification and regression tasks. • ε: a parameter to the epsilon-SVR model, applies to regression tasks. Defines the distance from true values within which no penalty is associated with predicted values. • ν-SVM Cost: penalty term for loss and applies only to regression tasks ν: a parameter to the ν-SVR model, applies to classification and regression tasks. An upper bound on the fraction of training errors and a lower bound of the fraction of support vectors. • Kernel is a function that transforms attribute space to a new feature space to fit the maximum-margin hyperplane, thus allowing the algorithm to create the model with Linear, Polynomial, RBF and Sigmoid kernels. Functions that specify the kernel are presented upon selecting them, and the constants involved are: • g for the gamma constant in kernel function (the recommended value is 1/k, where k is the number of the attributes, but since there may be no training set given to the widget the default is 0 and the user has to set this option manually), • c for the constant c0 in the kernel function (default 0), and • d for the degree of the kernel (default 3). • Set permitted deviation from the expected value in Numerical Tolerance. Tick the box next to Iteration Limit to set the maximum number of iterations permitted.
802
I. Zidane et al.
3.4 Gaussian Naive Bayes Naive Bayes classifiers are a family of simple probabilistic classifiers based on applying Bayes’ theorem with strong (naive) independence assumptions between the features. Bayes’ theorem (alternatively Bayes’ law or Bayes’ rule) describes the probability of an event, based on prior knowledge of conditions that might be related to the event. For example, if cancer is related to age, then, using Bayes’ theorem, a person’s age can be used to more accurately assess the probability that they have cancer, compared to the assessment of the probability of cancer made without knowledge of the person’s age. Naive Bayes is a simple technique for constructing classifiers: models that assign class labels to problem instances, represented as vectors of feature values, where the class labels are drawn from some finite set. It is not a single algorithm for training such classifiers, but a family of algorithms based on a common principle: all naive Bayes classifiers assume that the value of a particular feature is independent of the value of any other feature, given the class variable. For example, a fruit may be considered to be an apple if it is red, round, and about 10 cm in diameter. A naive Bayes classifier considers each of these features to contribute independently to the probability that this fruit is an apple, regardless of any possible correlations between the color, roundness, and diameter features. 3.5 Neural Network The Neural Network uses sklearn’s Multi-layer Perceptron algorithm that can learn non-linear models as well as linear (Fig. 8). Model parameters: Neurons per hidden layer: defined as the ith element represents the number of neurons in the ith hidden layer. E.g. a neural network with 3 layers can be defined as 2, 3, 2. Activation function for the hidden layer: Identity: no-op activation, useful to implement linear bottleneck Logistic: the logistic sigmoid function tanh: the hyperbolic tan function ReLu: the rectified linear unit function Solver for weight optimization: L-BFGS-B: an optimizer in the family of quasi-Newton methods SGD: stochastic gradient descent Adam: stochastic gradient-based optimizer Alpha: L2 penalty (regularization term) parameter Max iterations: maximum number of iterations
Epileptic Seizure Prediction
803
Fig. 8. Orange predictions for 5 models Workflow
Table 8. PERFORMANCE SCORES Model
AUC
CA
F1
Precision
Recall
Logistic Regression
1,000
1,000
1,000
1,000
1,000
kNN
0,999
0,925
0,919
0,932
0,925
Naive Bayes
0,967
0,903
0,909
0,927
0,903
Neural Network
1,000
1,000
1,000
1,000
1,000
SVM
0,987
0,938
0,941
0,950
0,938
Our paper targets in detecting seizures using 5 different machine learning algorithms. we used efficient and well-known supervised learning algorithms (Table 8). The most machine learning algorithms use statistics. According to the results obtained, we have observed that KNN is an algorithm closer to reality than the other algorithms, especially since we only used 150 classes which is a small number of classes.
804
I. Zidane et al.
Acknowledgments. This paper analyzes a large epileptic seizures database containing a lot of records EEGs signal processing despite of its complexity. This study succeeded thanks to the medical staff who had recorded the EEGs for 3 years and also the patients who contributed by their commitments without forgetting the staff of the university. Most of predictive methods use total seizures or cost-sensitive support vector machines without analyzing the other classes and features, thus will involve inaccurate predictions. Our method use all classes, features and parameters thanks to effective search easily and efficiently of pertinent and significant data using IA models. The prediction using deep learning algorithms, applied to 150 records in the epileptic dataset. The challenge for the future is to predict the epileptic seizure time before at least several minutes.
References 1. Scheffer, I.E., Berkovic, S., Capovilla, G., Connolly, M.B., French, J., et al.: ILAE classification of Theepilepsies: position paper of the ILAE commission for classification and terminology. Epilepsia 58, 512–521 (2017) 2. Kobayashi, K., James, C.J., Nakahori, T., Akiyama, T., Gotman, J.: Isolation of Epileptiform discharges from Unaveraged EEG by independent component analysis. Clin. Neurophysiol. 110(10), 1755–1763 (1999) 3. Ossadtchi, A., Baillet, S., Mosher, J.C., Thyerlei, D., Sutherling, W., Leahy, R.M.: Automated interictal spike detection and source localization in magnetoencephalography using independent components analysis and spatio-temporal clustering. Clin. Neurophysiol. 115(3), 508–522 (2004) 4. Williamson, J.R., Bliss, D.W., Browne, D.W., Narayanan, J.T.: Seizure prediction using EEG spatiotemporal correlation structure. Epilepsy Behav. 25(2), 230–238 (2012) 5. Chisci, L., et al.: Real-time Epileptic Seizure prediction using AR models and support vector machines. IEEE Trans. Biomed. Eng. 57(5), 1124–1132 (2010) 6. Rasekhi, J., Mollaei, M.R.K., Bandarabadi, M., Teixeira, C.A., Dourado, A.: Preprocessing effects of 22 linear univariate features on the performance of seizure prediction methods. J. Neurosci. Methods 217, 9–16 (2013) 7. Mirowski, P., Madhavan, D., LeCun, Y., Kuzniecky, R.: Classification of patterns of EEG synchronization for Seizure prediction. Clin. Neurophysiol. 120(11), 19271940 (2009) 8. Park, Y., Luo, L., Parhi, K.K., Netoff, T.: Seizure prediction with spectral power of EEG using cost-sensitive support vector machines. Epilepsia 52(10), 1761–1770 (2011) 9. Feldwisch-Drentrup, H., Schelter, B., Jachan, M., Nawrath, J., Timmer, J., Schulze-Bonhage, A.: Joining the benefits: combining epileptic Seizure prediction methods. Epilepsia 51, 1598– 1606 (2010) 10. Gadhoumi, K., Lina, J.-M., Gotman, J.: Seizure prediction in patients with mesial temporal lobe epilepsy using EEG measures of state similarity. Clin. Neurophysiol. 124(9), 17451754 (2013) 11. Paul, M., Lin, W., Lau, C.T., Lee, B.: Direct Intermode selection for H.264 video coding using phase correlation. IEEE Trans. Image Process. 20(2), 461–473 (2011) 12. Li, S., Zhou, W., Yuan, Q., Liu, Y.: Seizure prediction using spike rate of intracranial EEG. IEEE Trans. Neural Syst. Rehabil. 21(6), 880–886 (2013)
New Approach of 3D Protein Structure Superimposition: Case Study of “SARS-COV-2” and “SARS-COV” Nabil Aqili1(B) , Wajih Rhalem1,2(B) , Mohamed Zeriab Es-sadek1,2 , Hayat Sedrati3,4(B) , Najib alidrissi5 , Saïd Jidane2,6 , Imane Allali7 , Salsabil Hamdi8 , Zainab Elouafi11 , Nassim Kharmoum9 , Mostafa Ezziyani10 , Lahcen Belyamani2,6 , and Hassan Ghazal1,2,9,11(B) 1 E2SN Research Team, ENSAM, Mohammed V University in Rabat, Rabat, Morocco
[email protected], [email protected], [email protected] 2 Moroccan Society of Digital Health, Rabat, Morocco 3 National School of Computer Sciences and Systems Analysis, ICES Team, Mohammed V University in Rabat, Rabat, Morocco [email protected] 4 National School of Public Health, Rabat, Morocco 5 Department of Surgery, School of Medicine, Mohammed VI University of Health Sciences, Casablanca, Morocco [email protected] 6 Emergency Department, Mohammed V Military Hospital, Mohammed V University in Rabat, Rabat, Morocco 7 Laboratory of Biology of Human Pathologies, Faculty of Sciences, Mohammed V University in Rabat, Rabat, Morocco 8 Environmental Health Laboratory, Department of Research, Pasteur Institute Morocco, Casablanca, Morocco 9 National Centre for Scientific and Technical Research, Rabat, Morocco [email protected] 10 Faculty of Sciences and Techniques of Tangier Computer Sciences Department, University of Abdelmalek Essaadi, Tangier, Morocco [email protected] 11 Laboratory of Genomics and Bioinformatics, School of Pharmacy, Mohammed VI University of Health Sciences, Casablanca, Morocco
Abstract. COVID-19 is a disease caused by SARS-CoV-2. The entry of a virus into human cells is a critical phase in its infection. The binding of the spike protein of SARS-CoV-2 to Angiotensin converting enzyme 2 ACE2, an enzyme found on the surface of human cells, initiates the infection. Online software tools that overlay the three-dimensional structures of viruses, including SARS-CoV-2, address the problem of structure superposition by overlaying the ACE2 and spike complexes of the first protein on those of the second protein. In this work, overlaying the three-dimensional structures of viruses was addressed by superimposing ACE2 and then applying the resulting transformation from this superposition to the spike. Finally, the root mean square deviation RMSD was calculated. We used the discrete to continuous DTC algorithm to align the 3D © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 805–815, 2023. https://doi.org/10.1007/978-3-031-35248-5_71
806
N. Aqili et al. structures; the results from the DTC were compared to the results from TopMatch and SuperPose, which are online tools for aligning 3D structures. The obtained results prove that aligning each structure then combining them is the most suitable approach for properly studying the structural evolution of SARS-CoV-2. Keywords: Structure alignment algorithm · Discrete To Continuous · TopMatch · SuperPose · Coronaviruses · SARSCov2
1 Introduction Coronaviruses (COVs) are a wide range of viruses that can cause disease in animals and in humans. Among them are those that cause respiratory illnesses ranging from the common cold to more serious, life-threatening illnesses. Severe acute respiratory syndrome (SARS), Middle East respiratory syndrome (MERS), and the new SARS-CoV-2 are the most recent and thus well-known. The latter, which was recently discovered, causes COVID-19. Its most common symptoms are fever, fatigue, and dry cough. However, some people are infected without experiencing very mild symptoms. The coronaviruses are part of the order of Nidovirales, of the family of Coronaviridae and of the subfamily of Coronavirinae, which regroups 4 types: the first ones, alpha and beta, infect mammals; the remaining, gamma and delta, infect birds and mammals [1]. Coronaviruses are enveloped with a single-stranded positive ribonucleic acid RNA genome. Three viral proteins are anchored in the viral envelope: the spike protein (S), the membrane protein (M) and the envelope protein (E), in addition to the nucleocapsid protein (N). Proteins M and E are involved in viral assembly and secretion. Protein S gathers in trimers on the surface of virions and plays a key role in the entry of the virus into its target cell. It consists of two domains: the S1 domain responsible for binding the virus to its receptor and the S2 domain responsible for the fusion of the viral envelope with a cell membrane. Fusion is activated by cellular proteases by cleavage of protein S [2]. Similar to SARS-CoV and MERS-CoV, the recent SARS-CoV-2 belongs to the genus Beta-corona-virus [3]. The attachment of the virion to the host cell is initiated by the interactions between the RBD of protein S and its receptor, which determines the host spectrum and the tissue tropism of the coronavirus. SARS-CoV and human coronavirus (HCoV-NL63) interact with human angiotensin converting enzyme 2 (hACE2) for virus entry [4–6]. The surface of ACE2 contains two virus-binding hot spots that are essential for the binding of SARS-CoV. Several naturally selected mutations in the SARS-CoV RBM (receptor binding motif) surround these hot spots and regulate the infectivity, pathogenesis, and interspecies and interhuman transmissions of SARS-CoV [6, 7]. Structure comparison is the process of analyzing two or more structures looking for their similarities in 3D conformation. Structure analysis is a powerful tool that can provide significant information about fold classes and the function of an unknown protein from a known one based on their structural similarities. Structural comparisons are very useful in prediction, protein engineering, modeling and drug design. That is, the 3D structures provide us with information not available from sequence comparison.
New Approach of 3D Protein Structure Superimposition
807
The three-dimensional structure of SARS-CoV corresponds to the general shape of the protein observable at the molecular level. It describes the different interactions between the elements of the secondary structure. It is stabilized by a series of interactions, most often leading to the construction of the hydrophobic core, possibly with salt bonds, hydrogen bonds, disulfide bridges, or even posttranslational modifications. The skeleton structure consists of a repeating sequence of three atoms (nitrogen, alpha carbon and carbon). Of these three atoms, the alpha carbon (Cα) is particularly important because it is the center point of each amino acid residue in the protein. At the evolutionary level, structures are more conserved than biological sequences [8]. However, processing the 3D structure seems to be a very complicated task. In this study, we focused on the three-dimensional analysis of the SARS-CoV virus based on the spatial distribution of alpha carbon positions in 3D space and their relationship to the evolution of each type of coronavirus. For this reason, this study included SARS and SARS-CoV-2. Both viruses use ACE2 as an entry point into the cell.
2 Algorithms In structure alignment, the most commonly used techniques are based on the distance matrix. The concept of the digital addressable lighting interface DALI program is to construct a distance matrix by calculating all the distances between the C-alpha atoms and then applying dynamic programming to evaluate the similarity [9]. In [10], the combinatorial extension algorithm (CE) divides the structure of the protein into fragments. The final alignment is found through the search for an optimal path in a similarity matrix constructed from a series of aligned fragment pairs (AFPs). TopMatch [11] tries to resolve the alignment problem by generating several basic alignments using the diagonal deviation of the kth order, the Dk procedure, which yields a spectrum of deviations between two structures. Each particular Dk value measures the similarity of the diagonals in the distance matrices of the compared conformations [12]. SuperPose [13] uses a modified quaternion eigenvalue approach. It begins by aligning sequence pairs using the Neeldman-Wunsh algorithm [14] and a scoring matrix. Based on the resulting score, Superpose may perform a secondary structure alignment if the threshold exceeds 25%. Subsequently, Superpose generates a difference distance matrix between aligned alpha carbons. The last step in this process is the calculation of the RMSD, which is based on the root mean square deviation quaternion superposition algorithm. In this work, the discrete to continuous (DTC) [15] approach has been applied to protein structure comparison, considering this problem as a largest common point set (LCP) problem. The problem is to determine whether there is an allowed transformation T such that: T (Q) ⊂ P
(1)
where P is the model set (the target) and Q is the data set (the query). The DTC algorithm is a point pattern matching technique that has been shown to be useful in a variety of fields [16–18]. Instead of generating a distance matrix, the DTC uses an interpolation polynomial (cubic spline) to represent the model set. DTC does not
808
N. Aqili et al.
match the two point sets directly; it tries to bring one of them (Q) onto the continuous representation of the other (P). After this pretreatment, possible isomorphism is offered. Finally, this processing is achieved by the calculation of the RMSD.
3 Results To illustrate the utility of the proposed idea, we used two molecules downloaded from the RCSB protein data bank [19], under the accession number: 6M0J, 2AJF for Sars-Cov2 and Sars-Cov respectively, and shown in Table 1. We first aligned only ACE2 and then Spikes. After we aligned the complex of ACE2 and spikes using conventional methods. Finally, the findings of our proposed solution are reported. Table1. Accession IDs and name of the molecule complexes chosen for the study SARS-Cov-2
SARS-Cov
PDB ID
6M0J
2AJF
Name
Crystal structure of sars-cov-2 spike receptor-binding domain bound with ace2
Coronavirus spike receptor-binding domain complexed with its receptor
Molecule: 1
Angiotensin-converting enzyme 2
Angiotensin-converting enzyme 2
Chains
A
Chains
A, B
Sequence length
603
Sequence length
597
Molecule: 2
SARS-coronavirus spike protein
SARS-coronavirus spike protein
Chains
E
Chains
E, F
Sequence length
229
Sequence length
180
Each macromolecule is made up of two molecules, one of which is ACE2 and the other of which is the spike. It is worth noting that, in some cases, a molecule may contain more than one chain (duplication of the chains). For example, chains A and B in SARSCoV (2AJF) are identical; the same is true for the rest. In our experiments, only chains A and E from each macromolecule are used. Finally, Jmol [20] software is used to animate the results.
New Approach of 3D Protein Structure Superimposition
809
3.1 ACE2 Alignment In this first experience, we aligned ACE2 from SARS-CoV-2 (6m0J) and ACE2 from SARS (2ajf). In this case, the RMSD calculated with TopMatch and Superpose was equal to 0.57, while the RMSD measured with the DTC was equal to 0.56 (see Fig. 1). The aligned length (number of matched points) as calculated by TopMatch, Superpose, and DTC was 597, 597, and 596 respectively.
Fig. 1. Alignment of ACE2 from SARS and SARS-CoV-2 using the DTC approach
3.2 Spike Alignment In this second experiment, we aligned the spike from SARS-CoV-2 (6m0J) and the spike from SARS (2ajf). In this case, for TopMatch, the RMSD value was 0.93, and 1.66 for Superpose, while the RMSD measured with the DTC was equal to 0.78 (see Fig. 2). TopMatch, Superpose, and DTC calculated the aligned length as 163, 174, and 165, respectively.
810
N. Aqili et al.
Fig. 2. Alignment of ACE2 from SARS and SARS-CoV-2 using the DTC approach
3.3 Total Complex Alignment (ACE2+ SARS and ACE2+ SARS-COV2) In the third experiment, we aligned the entire complex composed of ACE2 and the spike SARS-CoV-2 (6m0J) and SARS (2ajf). The TopMatch RMSD values were 0.71 and 4.638 for Superpose, while the RMSD measured with the DTC was also equal to 0.71 (see Fig. 3). TopMatch and DTC assessed the aligned length to be 760 and 765, respectively. However, this desired chain cannot be specified by the Superpose tool.
New Approach of 3D Protein Structure Superimposition
811
Fig. 3. Alignment of ACE2+ SARS and SARS-CoV-2 (complex total)
3.4 Our Proposed Solution The last experience represents our proposed solution that consists of aligning ACE2. The resulting transformation of this alignment (ACE2 from SARS-CoV-2 and ACE2 from SARS) was applied to the spike. The measurement of the RMSD value, in line with the conventional technique, yielded the same result, namely, 0.71, using either the TopMatch tool or the DTC algorithm. Additionally, the aligned length measured using the DTC algorithm, according to the conventional method on the one hand and with our proposed solution on the other hand, was similar (i.e., 765 points). However, the RMSD value of the proposed solution measured with the DTC was 0.77 (see Fig. 4).
812
N. Aqili et al.
Fig. 4. Alignment of SARS and SARS-CoV-2 (proposed solution)
4 Discussion Understanding protein evolution and function is key to being able to characterize proteins and protein domains based on their structure and amino acid sequence. Many scientific topics of interest rely on being able to successfully establish the group of structures to which domains of a protein belong; this may be accomplished by protein structure alignment and comparison. Several approaches for protein structure alignment (PSA) have been established, each employing a unique set of methodologies. In this article, we offer a solution based on the data that aligning 3D structures is always impacted by “outliers”. Our proposed approach is to align the protein with the least evolutionary modification first and then apply the resulting transformation of this alignment to the protein that contain deformation or outliers. Using the DTC algorithm, in comparison with two structure alignment tools, TopMatch and SuperPose
New Approach of 3D Protein Structure Superimposition
813
we can conclude that the proposed approach is the best method for examining the threedimensional structures. In this coronavirus case study, ACE2 protein was first aligned, then the resulting transformation of this alignment was applied to the spike of Sars-Cov, all based on the spatial distribution of alpha carbon positions in 3D space and their relationship to the evolution of each coronavirus type. TopMatch has proven to be capable of computing the similarities of protein structures, whether single chain or multi chain [11]. The results produced with TopMatch methodology were consistent and coherent in terms of RMSD and aligned length measurements with SuperPose and DTC methods, with the exception of SuperPose in the examples of the Spike alignment and total complex alignment (ACE2+Spike). The RMSDs of the ACE2, SPIKE, and total complex (ACE2+Spike) of the two retrieved complexes 6M0J/2AJF were 0.57, 0.93, and 0.71, respectively. It was also found proficient at producing superpositions of 597, 163, and 760 matched pairs of sequence for ACE2, SPIKE, and total complex (ACE2+Spike) respectively. SuperPose had been historically found to be good at calculating protein superpositions using a modified quaternion approach [13]. SuperPose deviated from TopMatch and DTC in both spike alignment and total complex alignment (ACE2+Spike), with RMSDs of 1.66 and 4.638, respectively. These are extremely low accuracies, with poor RMSD values that preferably should always be not higher than 1.5 Angstrom. Furthermore, in the case of ACE2 alignment, SuperPose gave accurate results similar to TopMatch with 0,57 RMSD and 597 aligned pairs of sequences. The method that scored the highest in this work is the DTC method. It showed steady results very similar to the TopMatch tool, with an RMSD of 0.56, 0.78 and 0.71 for ACE2, SPIKE, and total complex (ACE2+Spike) respectively. The Aligned length measurements were 596, 165 and 765 for ACE2, SPIKE, and total complex (ACE2+Spike) alignment. The results of most of the alignment methods were well correlated between the number of aligned residues and the RMSD that were well balanced. Length dependency has long been known to affect the outcomes of alignment algorithms; the more residues there are to align, the higher the score is going to be. This can to some extent be offset by the fact that protein length tends to act to the detriment of RMSD, but not entirely [21]. However, our proposed solution scored a 0.77 RMSD and 765 aligned length measurements, where we aligned the ACE2 protein first and then the resulting transformation of this alignment was applied to the spike of Sars-Cov. All of this corroborated the predictions that alignment algorithms are sensitive to outlines. On that account, in the event of a Marco-molecule composed of two or more molecules, it is advised to align the patterns that are the most similar first, and then apply the transformation acquired to the remainder (the patterns that contain deformation or outliers). In the case of SARS-CoV2 and SARS-CoV viruses that interact with ACE2, it is better to align ACE2 and then apply the alignment’s transformation to the spike. In addition, the results of this study could suggest that DTC could be more reliable, as this algorithm is deterministic; it is certainly more reproducible and achieved more appropriate results than conventional methods (Table 2).
814
N. Aqili et al.
Table 2. RMSD score and aligned length measurements results for each method described above with ACE2 alignment, SPIKE alignment and total complex alignment (ACE2 + SPIKE) Tools
TopMatch
Values
RMSD
ALM
RMSD
SuperPose ALM
RMSD
DTC ALM
ACE2 alignment
0.57
597
0.57
597
0.56
596
SPIKE alignment
0.93
163
1.66
174
0.78
165
Total (ACE2+SPIKE) alignment
0.71
760
4.638
–
0.71
765
Proposed solution
–
–
–
–
0.77
765
5 Conclusion This study represents a new method to examine the three-dimensional structures of coronaviruses based on the spatial distribution of alpha carbon positions in 3D space and their relationship to the evolution of each type of coronavirus. The proposed method is adapted from the concept of shifting from discrete to continuous form (DTC) and consists to overlay the three-dimensional structures of viruses by first superimposing ACE2 and then applying the resulting transformation to the spike. Aligning the most similar protein patterns using the DTC algorithm yields more appropriate results than conventional methods for investigating structural similarity. The proposed approach could be effective in detecting SARS-CoV-2 residues required for the binding to the ACE2 cellular receptors. Our method is a good tool to be used to identify new compounds that bind to spike protein binding sites on the ACE2 receptor, which could be new drug leads preventing coronavirus entrance into human cells. Acknowledgments. This work was supported by the “Urgence COVID-19” fundraising campaign of the Institut Pasteur. This research is also supported through computational resources of HPC-MARWAN (www. marwan.ma/hpc) provided by the National Center for Scientific and Technical Research (CNRST), Rabat, Morocco.
References 1. Masters, P.S.: The molecular biology of coronaviruses. Adv. Virus Res. 66, 193–292 (2006) 2. Bonnin, A.: Characterization of human coronavirus 229E spike protein (2018). https://tel.arc hives-ouvertes.fr/tel-02275786 3. Lu, R., et al.: Genomic characterisation and epidemiology of 2019 novel coronavirus: implications for virus origins and receptor binding. Lancet Lond. Engl. 395, 565–574 (2020). https:// doi.org/10.1016/S0140-6736(20)30251-8 4. Hofmann, H., Pyrc, K., van der Hoek, L., Geier, M., Berkhout, B., Pöhlmann, S.: Human coronavirus NL63 employs the severe acute respiratory syndrome coronavirus receptor for cellular entry. Proc. Natl. Acad. Sci. 102, 7988–7993 (2005). https://doi.org/10.1073/pnas. 0409465102
New Approach of 3D Protein Structure Superimposition
815
5. Li, W., et al.: Angiotensin-converting enzyme 2 is a functional receptor for the SARS coronavirus. Nature 426, 450–454 (2003). https://doi.org/10.1038/nature02145 6. Li, F., Li, W., Farzan, M., Harrison, S.C.: Structure of SARS coronavirus spike receptorbinding domain complexed with receptor. Science 309, 1864–1868 (2005). https://doi.org/ 10.1126/science.1116480 7. Shang, J., et al.: Structural basis of receptor recognition by SARS-CoV-2. Nature 581, 221–224 (2020). https://doi.org/10.1038/s41586-020-2179-y 8. Koehl, P.: Protein structure similarities. Curr. Opin. Struct. Biol. 11, 348–353 (2001) 9. Holm, L., Rosenström, P.: Dali server: conservation mapping in 3D. Nucleic Acids Res. 38, W545–W549 (2010). https://doi.org/10.1093/nar/gkq366 10. Shindyalov, I.N., Bourne, P.E.: Protein structure alignment by incremental combinatorial extension (CE) of the optimal path. Protein Eng. Des. Sel. 11, 739–747 (1998). https://doi. org/10.1093/protein/11.9.739 11. Wiederstein, M., Sippl, M.J.: TopMatch-web: pairwise matching of large assemblies of protein and nucleic acid chains in 3D. Nucleic Acids Res. 48, W31–W35 (2020). https://doi.org/10. 1093/nar/gkaa366 12. Sippl, M.J.: On the problem of comparing protein structures. J. Mol. Biol. 156, 359–388 (1982). https://doi.org/10.1016/0022-2836(82)90334-5 13. Maiti, R., Van Domselaar, G.H., Zhang, H., Wishart, D.S.: SuperPose: a simple server for sophisticated structural superposition. Nucleic Acids Res. 32, W590–W594 (2004). https:// doi.org/10.1093/nar/gkh477 14. Needleman, S.B., Wunsch, C.D.: A general method applicable to the search for similarities in the amino acid sequence of two proteins. J. Mol. Biol. 48, 443–453 (1970) 15. Aqili, N., Raji, M., Jilbab, A., Chaouki, S., Hammouch, A.: PPM translation, rotation and scale in D-dimensional space by the discrete to continuous approach. Int. Rev. Comput. Softw. IRECOS 11, 270 (2016). https://doi.org/10.15866/irecos.v11i3.8746 16. Rhalem, W., et al.: An efficient and rapid method for detection of mutations in deoxyribonucleic acid-sequences. Int. J. Adv. Comput. Sci. Appl. 11, 278–286 (2020). https://doi.org/10. 14569/IJACSA.2020.0110438 17. Rhalem, W., Raji, M., Hammouch, A., Ghazal, H., El Mhamdi, J.: New algorithm for aligning biological data. In: Bhateja, V., Satapathy, S.C., Satori, H. (eds.) Embedded Systems and Artificial Intelligence. pp. 713–721. Springer, Singapore (2020). https://doi.org/10.1007/978981-15-0947-6_68 18. Rhalem, W., et al.: Novel alignment approach of DNA sequences. In: El Moussati, A., Kpalma, K., Ghaouth Belkasmi, M., Saber, M., Guégan, S. (eds.) SmartICT 2019. LNEE, vol. 684, pp. 489–497. Springer, Cham (2020). https://doi.org/10.1007/978-3-030-53187-4_53 19. Zardecki, C., Dutta, S., Goodsell, D.S., Voigt, M., Burley, S.K.: RCSB protein data bank: a resource for chemical, biochemical, and structural explorations of large and small biomolecules. J. Chem. Educ. 93, 569–575 (2016). https://doi.org/10.1021/acs.jchemed.5b0 0404 20. Jmol: an open-source Java viewer for chemical structures in 3D. http://jmol.sourceforge.net/ 21. Brown, P., Pullan, W., Yang, Y., Zhou, Y.: Fast and accurate non-sequential protein structure alignment using a new asymmetric linear sum assignment heuristic. Bioinformatics 32, 370– 377 (2016). https://doi.org/10.1093/bioinformatics/btv580
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application in Biomedicine Faoussi Mohamed1(B) , Tbib Bouazza2 , and Bounou Salim1 1 Euromed Research Center, School of Biomedtech Engineering, Université EUROMED de
Fès, Fez 3000, Morocco [email protected] 2 Nano-Sciences and Modeling Laboratory, Moulay Silman University, Khouribga, Morocco
Abstract. The use of effective contrast agents in magnetic resonance imaging (MRI) has become an international research challenge. The aim is to improve the diagnosis of some pathologies. The preparation of contrast agent by the solid-state method using efficient nanomaterials has attracted considerable attention due to its potential to interact with the physical stimulus, the objective is for the signal collected to be as intense as possible. Methods: This article proposes a simple method to prepare a nanomaterials based on pyrophosphates and doped with Gadolinium and Cerium from solid-state reagents. Firstly, the two phosphate nanomaterials (Pyrophosphate SrMP2 O7 ) based on Gadolinium and Cerium (SrMP2 O7 with M = Gd and Ce) were prepared at 1,000 °C to develop pure phases, capable of increasing the contrast of the signal observed by MRI. The characterization of these nanomaterials was mad by several techniques: X-ray diffraction (XRD) to determine the mesh parameters; UV-visible to interpret optical properties and electronic properties; Fourier transform infrared spectroscopy (FT-IR) to characterize the functional groups of SrGdP2 O7 and SrCeP2 O7 nanomaterials. Results: The Gap energies values of two nanomaterials showed that the SrCeP2 O7 has a better magnetic speed 1/T1 in comparison with the SrGdP2 O7 . Morever in T1 encoding, if the T1 relaxation time is short, the image is much clearer. However, T2 encoding implies that if the T2 relaxation time is short, the image is darker. Conclusion: Our nanomaterial based on Cerium SrCeP2 O7 have a better longitudinal relaxation. Then it can be an excellent positive contrast agent. Keywords: Gap energy · Medicine · Nanomaterials · Positive contrast agent · Pyrophosphate
1 Introduction Nanotechnologies constitute a field of multidisciplinary research and development, which is based on the crossing of several scientific disciplines such as mechanics, electronics, chemistry, and optics, biology while allowing the manipulation and characterization of matter at the nanometric scale. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 816–828, 2023. https://doi.org/10.1007/978-3-031-35248-5_72
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application
817
The study and modification of matter at the nanoscale gives unexpected results and properties that are often totally different from those of the same materials at the micro or macroscopic scale, in particular in terms of mechanical resistance, electrical conductivity, fluorescence and chemical reactivity. Indeed, nanotechnologies allow us to manufacture materials whose fundamental properties can be controlled [1]. Due to their varied and often unprecedented properties, the application of materials in nanomedicine presents itself as a transversal field that promises to revolutionize therapy and diagnostics in the health field. «Nano-theranostic» [2], by allowing the development of the principles of predictive medicine, personalized medicine and regenerative medicine [3]. In magnetic resonance medical imaging, some areas are difficult to visualize. Therefore, contrast agents needed to improve the diagnosis of some pathologies. The nanomaterials used to formulate these contrast agents must choose to interact with the physical stimulus so that the signal collected is as intense as possible. These agents must exhibit zero or at least low toxicity with respect to the actual benefit [4]. The goal of contrast agent injection is to accelerate the 1/T1 and 1/T2 magnetic relaxation speed of protons in water molecules, that is, to shorten the time during which the spins of these protons return to their initial state after excitation by the radiofrequency wave. This is what increases the contrast of the signal observed by MRI [5]. Studies have shown that Gadolinium Gd3+ ion is the element of choice for manufacturing MRI contrast agents [6]. Even today it is widely used in contrast agents. However, the high toxicity is the great inconvenience of this chemical element, this toxicity is due to its competition with the calcium ion (Ca2+ ) which is involved in multiple functions essential to the body: blood clotting, muscle contraction, nerve conduction, release of hormones. Researchers have shown that this toxicity can be considerably reduced by trapping it in ligand molecules: linear polyaminocarboxylates and macrocyclic. Gadolinium chelates are then obtained, they are also called contrastophores and many of which have been marketed [4]. However, other researchers have shown in a study, the existence of Gd3+ ions in skin biopsies of patients. Who underwent an MRI examination with injection of chelated Gadolinium in the months before the first symptoms, these ions would then precipitate in the form of Gadolinium phosphate, which would be phagocytosed by macrophages, which would then recruit circulating fibrocytes, triggering fibrosis [5]. Consequently, the competent bodies triggered the alerts. In the United States, the Food and Drug Administration (FDA) has issued an alert on the injection of all Gadolinium chelates, especially in patients with severe kidney failure; while in Europe, three of the least stable products (Omniscan®, Optimark® and Magnevist®) are contraindicated in these patients. The European Medicines Agency (EMEA) extends the prudence recommendation to the use of all other Gadolinium chelates in patients with severe kidney failure [7]. In addition, Cerium is the first element of the 17 lanthanides (rare earth elements) with a 4f orbital, which gives it physicochemical properties that are required in wide fields of application, such as: light, electricity, magnetism and other fields were witnessed [1, 8]. Indeed, the therapeutic applications of Cerium are numerous, thanks to its
818
F. Mohamed et al.
catalytic properties and its relevant pharmacological properties, for example, it is used as antiemetics, bacteriostats and antitumors [9]. Currently, efforts are focused on future applications of Cerium, in particular in nanomedicines [10, 11]. The objective of our study is to manufacture a stable and effective contrast agent based on Cerium. Indeed, we adopted a rational approach, starting from the solid method, of two phosphate nanomaterials (Pyrophosphate SrMP2 O7 ) based on Gadolinium and Cerium (SrMP2 O7 with M = Gd and Ce), to make a study of characteristics and comparison between the two nanomaterials.
2 Experimental 2.1 Synthesis of Materials Based on Gadolinium and Cerium SrMP2 O7 (M = Gd or Ce) For our study, the pyrophosphates SrGdP2 O7 and SrCeP2 O7 were synthesized by the solid-state method. The reagents chosen for the synthesis are: Gadolinium oxide (Gd2 O3 ) at 99.999%, 99.99% of SrCO3 (strontium carbonate) and ammonium phosphate (NH4 )2 HPO4 and Hydrated Cerium Acetate (C6 H9 CeO6 .H2 O) to 99.9% (Fig. 1). The synthesis reaction is as follows: SrCO3 + 0.5Gd2 O3 + 2(NH4 )2HPO4- →SrGdP2 O7 + CO2 + 3H2 O + 4NH3 + 0.25O2 SrCO3 + C6 H9 CeO6 .xH2 O + 2(NH4 )2 HPO4 →SrCeP2 O7 + CO2 + 2CH3 CO2 H + 2NH3 + 7.5H2
Fig. 1. Picture of the two nano-materials SrMP2 O7 (M = Gd or Ce) inside the alumina crucible at 25 °C.
The reagent mixtures in stoichiometric proportions being carefully ground in a mortar, in order to obtain a homogeneous mixture. Then introduced into a platinum crucible and placed in an oven to undergo a decomposition treatment: evaporation of structural water at 160 °C, then decomposition of the organic material at 450 °C (See Figs. 2 and 3). This decomposition follows a heat treatment according to the diagram below.
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application
819
SrCeP2O7 Almond color And SrGdP2O7 White color
Fig. 2. Picture of the two nano-materials SrMP2 O7 (M = Gd or Ce) at 300 °C.
Fig. 3. Picture of of the two nano-materials SrMP2 O7 (M = Gd or Ce) at 1000 °C.
Fig. 4. Reagent decomposition with heat treatment cycle
3 Results and Interpretation 3.1 X-Ray Diffraction The spectra of the nanomaterials were made at room temperature using a Philips Xpert MPD diffractometer, in CuKα radiation stepwise scanning mode, its step value is 0.02°. X-ray analysis of SrGdP2 O7 shows that this phase crystallizes in the monoclinic system (Space group: P2/m) with the following unit lattice parameters: a = 17.824
820
F. Mohamed et al.
Å; b = 3.0517 Å; c = 17.330 [5] Å. X-ray analysis of SrCeP2 O7 shows that this phase crystallizes in the monoclinic system (Space group: P2/m) with the following unit lattice parameters: a = 18.941 Å; b = 9.690 Å; c = 14.238 Å. The Rietveld coefficients obtained reveal a good agreement between the profiles of the observed and calculated diagrams. Figure 4 gives the X-ray diffraction spectrum, refined by the FullProf software of the two nanomaterials SrGdP2 O7 and SrCeP2 O7 (Fig. 5 and Table 1).
Fig. 5. X-ray Diffraction Spectrum SrGdP2O7 et SrCeP2O7
Table 1. Structural data of SrGdP2 O7 and SrCeP2 O7 nanomaterials Samples
SrGdP2 O7
SrCeP2 O7
Cristal system
Monoclinic
Monoclinic
S.Group
P 2/M
P 2/M
a(Å)
17.8243
18.9416
b(Å)
3.0517
9.6907 (continued)
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application
821
Table 1. (continued) Samples
SrGdP2 O7
SrCeP2 O7
c(Å)
17.3305
14.2384
v(Å)
907.2986
2489.1472
Alpha(°)
90.000
90.00
Beta(°)
107.750
96.078
Gama(°)
90.000
90.00
u
0.0100
0.0100
V
−0.0100
−0.0100
w
0.0050
0.00500
Rp
59.4
97.9
Rwp
62.5
61.5
Rexp
35.7
58.1
3.2 Optical properties of Pyrophosphates SrMP2 O7 with M = Gd and Ce Transmittance Spectra The pure pyrophosphates based on Gadolinium and Cerium (SrMP2 O7 with M = Ce and Gd) are presented in Fig. 6, they were analyzed by UV-Visibile absorption spectroscopy in order to demonstrate the absorption of Cerium and Gadolinium ions in this wavelength range. The Gadolinium spectrum is composed of a contribution centered at 310 nm. Our results are similar to a study by Shadab Ali and All [12]. Indeed, it is possible to distinguish the absorption of Gd ions, having an absorption band around 320 nm. This pyrophosphate SrGdP2 O7 has a transmittance greater than 35% in the visible range ranging from 400 nm to 600 nm. This band signifies an electronic transition of Gadolinium. We note that the SrGdP2 O7 nanomaterial is a semiconductor with a direct gap. We also have a transmittance equal to 40% in the near UV-visible-infrared region (from 400 to 850 nm). In addition, we have an absorption band which is due to the presence of free e-electron or that of conduction in the nanomaterial SrGdP2 O7 , which is naturally moves in all directions throughout the solid, under the effect of electromagnetic fields generated by other charged species. Indeed, this interaction of electromagnetic waves with matter can clearly explain the optical properties of the material. In the case of the SrCeP2 O7 nanomaterial, we have a wide electronic transition band; the transmittance is greater than 65% in the near UV-visible-infrared region (from 400 to 850 nm).
822
F. Mohamed et al.
Fig. 6. Transmittance spectra of SrMP2 O7 pyrophosphates with M = Ce and Gd
The Gap: To determine the optical band gap of our nanomaterials we used TAUC’s law and the Uv-visible spectrum. αhν = α0 (hν − Eg)n with n = 2 in the case (direct gap) With α: absorption coefficient, h: Planck constant, α0: constant; Eg: gap energy; Then the method consists in representing the (αhν)2 in function (hν) the value of the gap is obtained by the extrapolation of the linear part of the curve on the abscissa axis (hν), Fig. 7.
Fig. 7. Dependence (αhυ)2 of the nanomaterials SrGdP2 O7 and SrCeP2 O7 in terms of the photon energy in eV
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application
823
The Fig. 7 shows the curves of the nanomaterials SrGdP2O7 and SrCeP2O7. Their optical gap energy values are: – Egopt = 1.12 eV for the SrGdP2 O7 nanomaterial – Egopt = 1.06 eV for the SrCeP2 O7 nanomaterial. The Urbach Energy (Disorder): The energy of Urbach reflects the disorder state of the material; it is related to the absorption coefficient by the following expression: α = α 0 e(hν/Eu) With α: absorption coefficient and photon energy (hν) and Eu is energy of Urbach. Important Urbach energy values are shown in Fig. 8 and Table 2.
Fig. 8. Disorder by extrapolation from the variation of ln(α) in terms of hν for the both nanomaterials SrGdP2 O7 and SrCeP2 O7
Table 2. The gap energy and the Urbach energy value for the pyrophosphates SrGdP2O7 et SrCeP2O7 Pyrophosphates
SrGdP2 O7
SrCeP2 O7
Egopt (ev)
1.12
1.06
Eu (ev)
0.18
0.0
Ion Radius (pm)
94
101
824
F. Mohamed et al.
3.3 Infrared Spectroscopy The vibrational study by infrared spectroscopy of SrGdP2 O7 and SrCeP2 O7 nanomaterials provides structural information, in particular the identification of the different basic groups forming the phosphate material. In fact, the number and frequency bands distribution depend on the nature of local symmetry of the P2 O7 4− anion. The spectrum of SrGdP2 O7 and SrCeP2 O7 nanomaterials (Fig. 9) shows the existence of a band around 741.82 cm−1 for SrGdP2 O7 pyrophosphates and at 745.96 cm−1 for SrCeP2 O7 pyrophosphates attributed respectively to asymmetric stretching vibrations and symmetrical with the P-Ô-P bridge. These bands are characteristic of pyrophosphate groups (P2 O7 4− ). In the 450–1300 cm−1 range, frequencies related to symmetrical and antisymmetric vibration modes of end groups (PO3)2− have been highlighted. The bands around 400–700 cm−1 have been attributed to the deformation and tilting modes of the PO3 groups. In addition, the existence of frequencies Vs(PO3 ) in the infrared spectrum indicates that the ring [P2 O7 ] adopts a bent configuration (Table 3).
Fig. 9. Transmittance Spectra of SrGdP2 O7 and SrCeP2 O7 Nanomaterials Table 3. Allocation of bands to SrGdP2 O7 and SrCeP2 O7 nanomaterials Frequency value (cm−1 )
Transmi-ttance (%)
Assignment –
SrCeP2 O7
474,11
80
νs PO3
500,72
74
557,26
88,11
SrGdP2 O7
Frequency value (cm−1 )
Transmi-ttance (%)
Assignment
459,92
72
νs PO3
νs PO3
540,96
66,66
νs PO3
νs PO3
620,04
86,9
νs PO5
–
(continued)
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application
825
Table 3. (continued) Frequency value (cm−1 )
Transmi-ttance (%)
Assignment
Frequency value (cm−1 )
Transmi-ttance (%)
635,28 654,51
Assignment
92,1
νs PO3
745,96
94,3
νs P-O-P
98,41
νs PO3
954,22
87,56
νs PO7
741,82
96,38
νs P-O-P
975,65
76,56
νs PO3
950,43
82,85
νs PO3
1010,16
76,87
νas PO3
1007,81
86,23
νas PO3
1045,99
60,68
νas PO3
1056,43
87,75
νas PO3
1002,56
62,42
νas PO3
1097,18
87,95
νas PO3
1240,12
96,9
νas PO3
1181,67
91,95
νas PO3
–
–
–
1254,06
89,11
νas PO3
–
–
–
1212
97,22
νas PO3
–
–
–
4 Discussion There are two main families of contrast agents: positive contrast agents or T1 agents (longitudinal relaxation) and negative or T2 contrast agents (transverse relaxation) [4]. The quality of MRI imaging relies on the gray-scale coding of the relaxation time of water protons. Indeed, in T1 encoding, if the T1 relaxation time is short, the image is much clearer. However, T2 encoding implies that if the T2 relaxation time is short, the image is darker [5]. Moreover, in a study of quantum spin liquids [13], in the presence of a gap, the longitudinal relaxation time T1 follows an activation law (exponential function in terms of the gap energy): T1 ∝ exp Egopt /kB T With T: the temperature; KB : Boltzmann constant. In addition, we have demonstrated in the experimental part using UV-visible spectrometry that the gap energy of our SrCeP2 O7 nanomaterial (Egopt = 1.06 eV) is lower than that of the SrGdP2 O7 nanomaterial (Egopt = 1.12 eV). As a result, using the activation law and the gap energy, the SrCeP2 O7 nanomaterial has a reduced T1 relaxation time compared to the SrGdP2 O7 nanomaterial. Therefore, a better speed of magnetic relaxation 1/T1. Indeed, our nanomaterial SrCeP2 O7 , which is based on Cerium, will be a good positive contrast agent. We also noticed, in the X-ray diffraction characterization of the SrCeP2 O7 nanomaterial, that the shape of the spectrum peaks is very small, this indicates that the average grain size of our nanomaterial based on Cerium is very small; Indeed, the widths of the spectra are directly proportional to the size of the grains [14]. So a better ability to trap water molecules in the SrCeP2 O7 structure.
826
F. Mohamed et al.
All these elements revealed to us the idea of reinforcing this experimental result of positive contrast agent based on SrCeP2 O7 . Indeed, we thought to increase the local proton concentration by introducing a mechanism (Figs. 10 and 11) to our nanomaterial such that water can easily penetrate in its internal structure. The aim is to amplify the contrast of the observed signal.
Fig. 10. Mechanism Picture
Fig. 11. Reaction mechanism of water molecules with the Cerium complex
5 Conclusion In magnetic resonance imaging, particularly in the use of contrast agents, the American and European notified bodies have issued an alert on the injection of all Gadolinium chelates in patients with severe renal insufficiency, because of the high toxicity
Synthesis of New Phosphate Nanomaterials “Pyrophosphates of Cerium” Application
827
of this chemical element Gd3+ . Conversely, the relevant catalytic and pharmacological properties of Cerium (Ce2+ ) have made it the subject of several therapeutic applications. In our study, we prepared two phosphate nanomaterials: SrGdP2 O7 and SrCeP2 O7 by the solid-state reaction route. The objective is to synthesize an effective nanomaterial that could be a better positive contrast agent based on Cerium. The Gap energies values of two nanomaterials showed that the SrCeP2 O7 has a better magnetic speed 1/T1 in comparison with the SrGdP2 O7 , so a better longitudinal relaxation. Therefore, our nanomaterial based on Cerium SrCeP2 O7 will be a good positive contrast agent.
References 1. Bazhukova, I.N., Sokovnin, S.Y., Ilves, V.G., Myshkina, A.V., Vazirov, R.A., Pizurova, N., et al.: Luminescence and optical properties of cerium oxide nanoparticles. Opt. Mater. (AMST) [Internet]. 92, 136–142 (2019). https://linkinghub.elsevier.com/retrieve/pii/S09253 46719302502 2. Nicolas, L., Leclerc, P., Guilloteau, D., Lebtahi, R.: Médicaments radiopharmaceutiques: du diagnostic au théranostique des tumeurs neuroendocrines. Médecine Nucléaire 45(3), 127–134 (2021) 3. Noury, M., Lafontaine, C.: De la nanomédecine à la nanosanté : vers un nouveau paradigme biomédical. Socio-anthropologie [Internet]. (29), 13–35 (2014). http://journals.openedition. org/socio-anthropologie/1635 4. Tsapis, N.: Agents de contraste pour l’imagerie médicale. médecine/sciences [Internet], vol. 33(1), pp. 18–24, 25 January 2017. http://www.medecinesciences.org/https://doi.org/ 10.1051/medsci/20173301004 5. Port, M.: Chapitre 9 : Les agents de contraste dans l’imagerie par résonance magnétique, pour le diagnostic médical. In: La chimie et la santé [Internet], pp. 153–168. EDP Sciences (2020). https://www.degruyter.com/document/doi/https://doi.org/10.1051/978-2-7598-09349.c012/html 6. Hermann, P., Kotek, J., Kubíˇcek, V., Lukeš, I.: Gadolinium(iii) complexes as MRI contrast agents: ligand design and properties of the complexes. Dalton Trans. (23), 3027 (2008). http:// xlink.rsc.org/?DOI=b719704g 7. Rasschaert, M.: Capture cérébrale de chélates de gadolinium: imagerie multimodale et analyse des conséquences neurotoxicologiques. Université Paris-Saclay (2019) 8. Zhang, S., Zhao, D. (eds.): Advances in Magnetic Materials: Processing, Properties, and Performance, p. 758. CRC Press, Boca Raton (2017) 9. Zambon, A., Malavasi, G., Pallini, A., Fraulini, F., Lusvardi, G.: Cerium containing bioactive glasses: a review. ACS Biomater. Sci. Eng. 7(9), 4388–4401 (2021). https://pubs.acs.org/doi/ https://doi.org/10.1021/acsbiomaterials.1c00414 10. Elayakumar, K., Dinesh, A., Manikandan, A., Palanivelu, M., Kavitha, G., Prakash, S., et al.: Structural, morphological, enhanced magnetic properties and antibacterial bio-medical activity of rare earth element (REE) cerium (Ce3+) doped CoFe2O4 nanoparticles. J. Magn. Magn. Mater. 476, 157–165 (2019). https://linkinghub.elsevier.com/retrieve/pii/S03048853 18326787 11. Li, H., Xia, P., Pan, S., Qi, Z., Fu, C., Yu, Z., et al.: The advances of Ceria nanoparticles for biomedical applications in orthopaedics. Int. J. Nanomed. 15, 7199– 7214 (2020). https://www.dovepress.com/the-advances-of-ceria-nanoparticles-for-biomed ical-applications-in-ort-peer-reviewed-article-IJN
828
F. Mohamed et al.
12. Khan, S.A., Gambhir, S., Ahmad, A.: Extracellular biosynthesis of gadolinium oxide (Gd2O3 ) nanoparticles, their biodistribution and bioconjugation with the chemically modified anticancer drug taxol. Beilstein J. Nanotechnol. 5, 249–257 (2014) 13. Bert, F., Mendels, P., Cépas, O., Lhuillier, C.: Quand la frustration rend plus dynamique: les liquides de spins quantiques. Reflets la Phys. 37, 4–11 (2013) 14. Eberhart, J.-P.: Analyse structurale et chimique des matériaux. Dunod, D. (ed.), Du. Paris, Paris, p. IV-614 (1997)
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization and Advanced Structure Study Application: Improve Mechanical Strength of Implant and the Osteointegration Process Faoussi Mohamed1(B) , Tbib Bouazza2 , Zakaria Kbiri3 , and Bounou Salim1 1 Euromed Research Center, School of Biomedtech Engineering, Université EUROMED de
Fès, Fez 3000, Morocco [email protected] 2 Nano-Sciences and Modeling Laboratory, Moulay Silman University, Khouribga, Morocco 3 Scientific Institute, University of Mohammed V, Av. Ibn Batouta, Agdal, BP 703, Rabat, Morocco
Abstract. Obtaining and maintaining implant stability are two essential conditions for the long-term functional success of bone-anchored prostheses. However, this initial stability also depends on other factors such as: the geometric shape of the implant, the clinician and the surgical protocol. After the primary healing, we have the secondary stability which is determined by the biological response to the surgical trauma, the patient’s state of health, the healing conditions, the material of which the implant is made and its biocompatibility (physical, chemical properties and mechanical). In this work we synthesized a new nanomaterial “Niobium Triphosphate NbP3 O10 ” by the solid-state method according to a well-defined thermal cycle (1000 °C for 6 h). The objective is to make an advanced structural study of our nanomaterial in order to identify its physical characteristics. The purpose of our nanomaterial is to increase the mechanical resistance of our implant and improve bone healing by a surface coating. The characterization and identification of Niobium Triphosphate NbP3 O10 was validated by three techniques: X-Ray diffraction; FTIR Spectroscopy; X-ray fluorescence Spectrometry. A finite element modeling is made on Catia V5, by applying Von Mises constraints in order to calculate the deformation energy. Keywords: Osseo-integration · Orthopedic · Nanomaterials · Implant Stability · Triphosphate · Mechanical Strength
1 Introduction Nowadays, osseointegration is widely used in dental, orthopedic, maxillofacial and cancer surgery where defects pose serious functional and aesthetic problems. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 829–844, 2023. https://doi.org/10.1007/978-3-031-35248-5_73
830
F. Mohamed et al.
The concept of osseointegration was defined by Per Ingvar Brånemark in 1965 as being a direct bone apposition on the implant surface [1], subsequently osseointegration took the name of “functional ankylosis” [2]. It is characterized by direct structural and functional coaptation between living bone and the implant surface [3]. It is the result of direct primary bone regeneration on the implant. This process is reflected in two phases, or by two complementary stabilities: immediate stability of the implant after its surgical placement, this corresponds to “primary” or “mechanical” stability; this phase gradually decreases to be replaced by “secondary” or “biological” stability, which is determined by a well-defined biological process (See Fig. 1).
Fig. 1. Diagram of the progressive decrease in primary stability, responsible for the immobility of the implant and the increase in biological stability
1.1 Mechanical Stability (Primary Stability) Primary stability represents the level of immobility of an implant after its surgical placement. It is a mechanistic criterion, which promotes bone healing and osseointegration. In the absence of satisfactory primary stability, the bone-implant interface may be the site of persistent micromovements. When they are too great, they lead to fibro-integration of the implant, synonymous with implant failure. In this case, we have filed a national invention patent No. 41535, with the aim of improving primary stability. This work was accepted and approved by the Patent Department of OMPIC Morocco, on 16 May, 2018. 1.2 Secondary Stability (Biological Stability) Secondary stability is a biological response to surgical trauma, through bone regeneration and remodeling. This bone regeneration takes place in several stages (See Fig. 2). – Initially, the alliance between the proteins of the inflammatory response (Ex: fibrogens) and the blood platelets, forms a network of fibrin, on which the cells necessary for bone remodeling can move.
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization
831
Fig. 2. The different stages of biological stability
– Then, chemical mediators are released, initiating a cascade of cellular activations starting with macrophages and neutrophils, which are responsible for cleaning the site, suppressing pathogens and managing inflammatory metabolism. – The previously formed fibrin network also allows mesenchymal stem cells (MSC) to migrate to the site. These cells differentiate into osteoblasts, which will gradually replace the fibrin matrix with a collagen bone matrix. Calcium (Ca) and phosphate (P) ions are deposited on this collagen matrix to form hydroxyapatite (HA) crystals – Finally, the osteoclasts will remodel this newly synthesized bone tissue and thus initiate a remodeling, in association with the osteoblasts which will remove small imperfections by filling in these resorptions to make the mature bone more resistant [4]. To improve the bone healing, we have several determining factors that can lead to the success or failure of the osseointegration process, they are divided into three categories: – Patient-related factors: the quality and quantity of a patient’s soft and hard tissues are directly related to their state of health (cardiovascular disorders, endocrine disorders, etc.). Lifestyle habits (sport, alcohol, tobacco, etc.) are also factors that have a great influence on the biological response of the human body, following an implantology intervention. – Factors related to the surgical technique: today, the surgeon has different surgical techniques to optimize primary stability, either by using screw systems, or systems with skeletal integration thanks to bone growth in a microporous structure of the implant. – Factors related to the implant: mechanical strength, biocompatibility and surface condition of the implant. In this work we are interested in the factors related to the implant, in particular the resistance to deformation and the surface condition, because they can significantly influence the success of the osseointegration process [5].
832
F. Mohamed et al.
Indeed, the surface condition determines the possibility for the blood to cover the material very quickly, inducing good bone healing. In addition, the wettability (wetting) designates on the one hand, the form that the blood takes on the surface of the implant (static wetting), and on the other hand, the behavior of the blood in contact with this surface. These behaviors arise from the interactions, which are determined by the surface energy. Therefore, the higher this energy, the greater the wettability. Scientific research has shown that several parameters influence the surface energy as well as the wettability of the implant, such as: the roughness and the composition of the biomaterial used. Indeed, these authors have confirmed the link between the power of wettability and the adhesion of bone cells [6–8]. In this article we worked on the identification of a new nanomaterial “niobium triphosphate (NbP3 O10 )”. Its characterization is made by three techniques: X-Ray diffraction, FT-IR Spectroscopy and X-ray Fluorescence Spectrometry. The usefulness of our nanomaterial is to promote the onsteointegration process thanks to an implant surface coating. In fact, bone contains about 85% of the body’s phosphate. It intervenes to promote thermodynamically the biochemical reactions of bone anabolism. In addition, Niobium (Nb) is an alloying agent that imparts unique properties to materials to which it is added. Indeed, the addition of niobium to titanium gives it undeniable advantages, at the medical, economic and environmental level, because it increases the mechanical resistance of our implant, lightens it and makes it hypoallergenic. Thanks to its exceptional properties.
2 Experimental 2.1 Synthesis of Niobium Triphosphates NbP3 O10 The precursors selected for synthesis were 99.999% Nb2O5 and 99.999% (NH4)2HPO4 (Aldrich). Niobium triphosphates were obtained by solid route according to the following reaction. 0.5Nb2O5 + 3(NH4)2HPO4 → NbP3O10 + xCO2 + yNO2 + zH2O + zO2 The stoichiometric proportions of the reactant mixtures being thoroughly ground in a mortar in order to obtain a homogeneous mixture facilitating vitrification are then introduced into a platinum crucible and then placed in an oven to undergo a decomposition treatment before melting. This preliminary cycle makes it possible to ensure the decomposition of the reactants and the gaseous releases occurring during the preceding reaction, namely H2O, NH3 from 155 °C and NO2 at 444 °C. (Figs. 3, 4, 5, 6, 7 and 8) 2.2 Characterization of Niobium Triphosphates NbP3 O10 a) X-Ray diffraction study X-ray powder pattern was carried out at room temperature using a diffractometer in the step scan mode (CuKα radiation, at a step value of 0.02°). X-ray analysis of the
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization
833
Fig. 3. Heat treatment cycle allowing the decomposition of the reagents
Fig. 4. Photograph of NbP3 O10 phosphates inside the alumina crucible at room temperature (25 °C)
Fig. 5. Photograph of NbP3 O10 phosphates inside the alumina crucible at 400 °C
834
F. Mohamed et al.
Fig. 6. Photograph of NbP3 O10 phosphates inside the alumina crucible at 1000 °C before grinding
Fig. 7. Photograph of NbP3 O10 phosphates inside the alumina crucible at 1000 °C after grinding
triphosphate de Niobium NbP3 O10 shows that this phase crystallizes in the Monoclinic system (Space Group: P 2/m) with the following unit cell parameters: a = 13.036 Å, b = 11.3182 Å, c = 12.539 Å, α = γ = 90°, β = 117.903° and direct Cell Volume = 892.546 Å3 . For the sample a major peak is observed at about 20.86°. The obtained Rietveld coefficients are indication of a good concordance between the observed and calculated diagram profiles. The Fig. 1 gives the observed, calculated and difference powder XRD profiles of triphosphate de Niobium NbP3 O10 . The Bragg positions are also indicated. Results of the structural refinement of triphosphate de Niobium NbP3 O10 are given in Table 1 (Fig. 8).
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization
835
Fig. 8. Observed, calculated and difference powder XRD profiles of Niobium NbP3 O10
The measurements are performed on (112) diffraction peak. The formed crystallite size of NbP3 O10 is calculated from the full width at half maximum (FWHM), of the (112) peak using Debye-Scherrer formula [9]. D = 0.9 ∗ λ/(βcosθ ) where D is the average crystallite size, β is the FWHM, λ is the wavelength of incident X-ray and θ is the angle of diffraction. The NbP3 O10 nanoparticles size was obtained in the above XRD pattern as 0.04 μm (Figs. 9, 10 and Table 3). b) FT-IR Spectroscopy The vibrational study by infrared spectroscopy of triphosphates NbP3 O10 allows having a very fast structural information, in particular the identification of the various basic groupings forming phosphatic material. Indeed, the number and the distribution of the bands frequencies depend on local symmetry nature of the P3 O10 5− anion. The spectrum of Niobium NbP3 O10 compound (Fig. 2) shows the existence of band around 954,5 cm−1 assigned with symmetric stretching vibration of the P-Ô-P Bridge. These bands are characteristic of triphosphates groups (P3 O10 5− ) In the 975–1300 cm−1 domain [10, 11], frequencies related to the symmetric and antisymmetric vibration modes of terminal (PO3 )2− groups have been evidenced. The bands around 400–700 cm−1 were assigned to the deformation and rocking modes of PO3 groups [12]. Furthermore, the existence of νs (PO3 ) frequencies in the infrared spectrum indicates that [P3 O10 ] ring adopts a bent configuration (Fig. 11 and Table 4).
836
F. Mohamed et al.
Fig. 9. Rietveld refinement spectrum of NbP3 O10 phase Table 1. Results of the Rietveld refinement of NbP3 O10 phase Formula
NbP3 O10
Crystal system, S.G
Monoclinic, P 2/m
a (Å)
12.6557
b (Å)
8.778
c (Å)
8.2890
α = γ = (°)
90°,
β=
112.148
Volume(Å)
852.9109 Å3
D(μm)
0.04
Rwp: Rexp:
23.66
X2
25.45
U
−0.628621
V
0.567999
W
0.008911
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization
837
Table 2. The atomic positions of NbP3 O10 Atoms
x
y
z
P1
0.08391
0.52639
0.35389
P2
0.06693
0.66691
0.44693
O1
0.13180
0.73996
0.54226
O2
−0.09692
0.37166
−0.31623
O3
1.03124
0.22189
0.45211
O4
0.08568
0.29053
0.44544
O5
0.30278
0.68424
0.65010
Nb1
0.00000
0.00000
0.00000
Nb2
0.33180
0.54569
0.30643
Nb3
0.24892
0.95946
0.36682
Fig. 10. Represent the cell of the structure NbP3 O10
c) X-ray fluorescence Spectrometry The triphosphate of Niobium NbP3 O10 was characterize by the X-ray fluorescence Fig. 2, the optical phenomena are present, all the objects are characterized by the fluorescence, the fluorescence it is an optical phenomenon in numerous applications this phenomenon based on the emission of the light by a seen material that it was enlightened, we are interested in the specters of absorption, wavelengths are absorbed by the material, it absorbs many lights when we enlightened the material by the fluorescence we brought the energy of the materials which passed in an excited state, I think that this state is very unstable, on the other hand, the incited electrons are going to fall again on a fundamental state with the emission of the light, just a little energy is lost by said no transitions radioactive, the photon emitted by the possessed material less energy than the photon which served to excite, this observation participated in the mechanism of interaction between the light and the material, the transfers of the energy between the
838
F. Mohamed et al. Table 3. Represents the values of energy during network vibration using Jmol Time
Energy
10
9.883367
20
0.127799
30
−1.093610
40
−1.988347
50
−2.655187
60
−3.158434
70
−3.541973
80
−3.836669
90
−4.060002
100
−4.219768
Fig. 11. The transmission spectra of Triphosphates NbP3O10
light and the atoms of the material are made by discreet jump and its discreet energy, it is the science of the quantum mechanics. The specter of the X-rays accumulated during this process reveals a number of characteristic peaks. The energies of peaks allow us to identify the present elements in the sample whereas the intensities of peaks supply the relative concentration. As these transitions concern electrons strongly bound of the internal layers, the energies of the levels, and consequently the energies of the transitions, are essentially independent from the chemical state of the atom. The transitions between outer layers can have relatively big variations of the energies of emission, due to the effects in the band of valence, and thus by generating information of diverse nature. The same information is contained in the transitions X which have their origin in outer layers. The relatively high energy of the X-rays of emission allowed an actual analysis in volume of material, because the energy radiation has a bigger probability of transmission through the material without being absorbed. Several elements are present in the triphosphate of Niobium NbP3 O10 , who translate the relative absorption, these a
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization
839
Table 4. Summarizes the band assignments for NbP3 O10 Frequency value (cm−1 )
Transmittance (%)
Assignment
2945 1648 1230,20 1082,65
91,34 95,90 82,97 75,20
νas PO3
954,50
56,54
νs PO3
954,50
56,54
νas P-O-P
623,24 567,9 488,6
82,71 79,16 60,10
δPO3 & ρPO3
* ν = stretching vibration, * νs = symmetric stretching
vibration, ν* as = asymmetric stretching vibration and * δ = wagging
linear combination of the concentrations as to present them in the Table 2. Both major elements are analyzed (Nb and P). Seventeen elements in tracks usually measured are it from pastilles. For other elements in tracks, in particular the elements of transition, spectral interferences can limit the dosage to samples presenting contents relatively more important. For elements in track, the limit of detection is generally 5 ppm, not taken into account here; elements in the Table 2 are measurable with nearby limits of analysis of 10 ppm. All the values quoted here correspond to made measures “in routine” with the standard preparation and could be improved by changing the analytical process, or the treatment according to matrices, elements which we want to favour (Fig. 12 and Table 5).
Fig. 12. The spectrum of x-ray fluorescence of triphosphate of Niobium NbP3O10
840
F. Mohamed et al. Table 5. Results of the XRF of NbP3 O10 phase Compound
Conc
Unit
0Mg
324,7
ppm
Al
0,185
%
Si
0,595
%
P Cl K Ca
10,8 0,387 922,8 6,168
% % ppm %
Ti
202,2
ppm
Cr
226,5
ppm
Fe
0,119
%
Cu
172,9
ppm
Zn
201,3
ppm
Sr Zr
0,135 391,7
% ppm
Nb
15,928
%
Ag
0,135
%
Ce Gd Re
331,9 0,481 18,3
Ppm % ppm
3 System Design, Mesh and Finite Element Modeling (FEM) a) Industrial Design In order to represent the mechanical characteristics of our implant, we drew on Catia V5 the real dimensions of an osteointegrated orthopedic implant. The study of the mechanical resistance of our implant, in which we apply the new nanomaterial “triphosphate of niobium”, is made by finite element modeling, by applying Von Mises constraints. The goal is to calculate the strain energy. We applied an embedding on the base of the implant plus a compression loading on the other side which will be in contact with the prosthetic foot. The compressive force applied is of the order of 300 N/m2 (see Fig. 13).
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization
841
Fig. 13. Drawing of the sample to be characterized with mechanical compressions
Structure Calculation: Number of nodes: 12208 Number of elements: 6480 Number of D.D.L: 36627 Number of contact relationships: 0 Number of kinematic relationships: 6 Number of coefficients: 339 b) Mesh The mesh chosen is of fine size at the millimeter scale, also we applied parabolic elements to provide reliable results. The results of these parameters are given by the following Fig. 14.
Fig. 14. Deformed mesh of our implant under mechanical loading
842
F. Mohamed et al.
c) Finite Element Modeling (FEM): Von Mises Constraints The Von Mises criterion is an energy criterion, it means the elastic strain energy, its mathematical formula in tension-compression is written (Fig. 15 and Table 6): U = 1 2σ ε σ: Constraint; ε: Deformation or Relative elongation
Fig. 15. Finite Element Modeling (FEM): Von Mises Constraints on Catia V5
Table 6. Balance Results Components
Strengths Applied
Reactions
Residues
Error Relative
Fx (N)
1,02E−13
−7,62E−12
−7,52E−12
1,06E−09
Fy (N)
−1,08E−13
−7,99E−12
−8,10E−12
1,15E−09
Fz (N)
−1,04E+03
1,04E+03
2,26E−10
3,20E−08
Mx (Nxm)
1,48E−14
3,23E−13
3,38E−13
3,42E−10
My (Nxm)
1,52E−14
−3,91E−13
−3,76E−13
3,80E−10
Mz (Nxm)
2,83E−15
8,25E−14
8,53E−14
8,62E−11
Results Strain energy (Implant with Triphosphate of Niobium NbP3 O10 ) = 2.795e−011 J. Strain energy (Implant with Titanium) = 2.501e−011 J. Discussion The results of our finite element modeling showed a difference in elastic strain energy between our nanomaterial “Triphosphate of Niobium NbP3 O10 ” and pure titanium.
Synthesis of a New Nanomaterial “Triphosphate of Niobium” Characterization
843
Indeed, to deform an implant made of our alloy it is necessary to provide more energy in comparison with titanium. This shows that our nanomaterial improves the mechanical strength of the orthopedic implant. Conclusion Our New triphosphates of Niobium NbP3 O10 powder is prepared by a solid-state method. This sample is thermally treated at 1000 °C for 6 h to get monazite type structure with orthorhombic system. The phase, nanomaterial and optical properties of triphosphates of Niobium sample is characterized by X-ray diffraction patterns, Fourier transform infrared spectra and fluorescence X (SFX). Unit cell parameters: a = 13.036 Å, b = 11.3182 Å, c = 12.539(5) Å, α = γ = 90°, β = 117.903° and The mean crystallite size calculate from the XRD pattern has been found to be 0.04 μm. Fourier transform infrared spectra confirms the presence of characteristic bands from P3 O10 phosphate group. Indeed, we have developed an identification map of our nanomaterial. These elements are potential for our second step which is the modeling of an orthopedic implant on CAD software, in order to verify these mechanical characteristics (Strain energy, Young’s modulus, Poisson’s ratio and its elastic limit). The objective is to compare the new nanomaterial with the titanium in order to validate our initial hypothesis, which is the improvement of the mechanical resistance.
References 1. Brånemark, P.I., Hansson, B.O., Adell, R., Breine, U., Lindström, J., O Hallén, A.O.: Osseointegrated implants in the treatment of the edentulous jaw. Experience from a 10-year period. Scand. J. Plast. Reconstr. Sug. 16, 1–132 (1977) 2. Schroeder, A.; Pohler, O.; Sutter F.: Tissue reaction to an implant of a titanium hollow cylinder with a titanium surface spray layer. Schweizerische Monatsschrift fur Zahnheilkd, pp. 713– 727 (1976) 3. Listgarten, M.A., Lang, N.P., Schroeder, H.E., Schroeder, A.: Periodontal tissues and their counterparts around endosseous implants. Clin. Oral Implants Res. 2(3), 1–19. http://doi. wiley.com/10.1034/j.1600-0501.1991.020309.x 4. Boyan, B.D., Cheng, A., Olivares-Navarrete, R., Schwartz, Z.: Implant surface design regulates mesenchymal stem cell differentiation and maturation. Adv. Dent. Res. 28, 10–17 (2016). http://journals.sagepub.com/doi/10.1177/0022034515624444 5. Smeets, R., et al.: Impact of dental implant surface modifications on osseointegration. Biomed. Res. Int. 2016, 1–16 (2016). http://www.hindawi.com/journals/bmri/2016/6285620/ 6. Buser, D., et al.: Enhanced bone apposition to a chemically modified SLA titanium surface. J. Dent. Res. 83(7), 529–533 (2004). http://journals.sagepub.com/doi/10.1177/154405910408 300704 7. Zhao, G., et al.: High surface energy enhances cell response to titanium substrate microstructure. J Biomed Mater Res Part A [Internet]. 74A(1), 49–58 (2005). https://onlinelibrary.wiley. com/doi/10.1002/jbm.a.30320 8. Schwarz, F., Herten, M., Sager, M., Wieland, M., Dard, M., Becker, J.: Histological and immunohistochemical analysis of initial and early subepithelial connective tissue attachment at chemically modified and conventional SLA®titanium implants. A pilot study in dogs. Clin. Oral. Investig. 11(3), 245–255 (2007). https://doi.org/10.1007/s00784-007-0110-7 9. Vinila, V.S., et al.: Ray diffraction analysis of nano crystalline ceramic PbBaTiO3. Cryst. Struct. Theory Appl. 3 (2014)
844
F. Mohamed et al.
10. Mohamed El Masloumia, C., Inhar Imaza, J.-P.C., Videaua, J.-J., Couzi, M., Mesnaoui, M., Maazaz, M.: Synthesis, crystal structure and vibrational spectra characterization of MI La(PO3)4 (MI ¼ Na, Ag). J Solid State Chem. 178, 3581–3588 (2005) 11. Krishna Bharat, L., Jeon, Y.I., Yu, J.S.: Synthesis and luminescent properties of trivalent rare-earth (Eu3+, Tb3+) ions doped nanocrystalline AgLa(PO3)4 polyphosphates. J Alloys Compd. 614, 443–447 (2014). https://linkinghub.elsevier.com/retrieve/pii/S09258388 14014649 12. Dardar, F.E., et al.: Synthesis, structural characterization and ionic conductivity of mixed alkali titanium phosphate glasses. Mediterr. J. Chem. 7(5), 328–336 (2018). http://medjchem-v3. azurewebsites.net/index.php/medjchem/article/view/790
Extraction and Characterization of New Cellulosic Fibers from Moroccan Mallow Stem and Comparison with Other Naturel Fibers Youssef El Omari(B) , Bouazza Tbib, Mohammed Eddya, Zakaria Kbiri, and Khalil El-Hami Scientific Institute, University of Mohammed V, BP 703, Av. Ibn Batouta, Agdal, Rabat, Morocco [email protected]
Abstract. Natural fibers are one of the good alternative sources for replacing synthetic fiber and reinforcing polymer matrices. Thanks to their eco-friendly nature. This investigation deals with the extraction and characterization of new natural fibers from the Moroccan mallow plant stem. The physicochemical, thermal, and mechanical properties of Moroccan Mallow Fibers (MMFs) were reported and compared with other natural fibers for the first time. Fourier transform-infrared spectroscopy, X-ray fluorescence (XRF) spectroscopy and X-Ray Diffraction, confirmed that MMFs are rich in cellulose content. Our object focused in the evolution of crystallinity index (CrI). We compared their properties to them of corn fibers. Keywords: Mallow · XRD · XRF · Lead · Rietveld refinement · FTIR · Crystallinity index
1 Introduction In the last few decades, the scientists and industry are given more important to the research and development of composite materials based on organic-materials originally of renewable sources [1]. The natural fibers were replacing synthetic fibers due to their eco-friendly behaviour, also to its non-toxicity, biodegradability and of their low cost and low density [2]. Cellulose is an organic compound with the formula (C6 H10 O5 )n , a polysaccharide consisting of a linear chain of several hundred to many thousands of β(1, 4) linked D-glucose units. In nature, cellulose chains are grouped together in order to form a compact microfibrils [3], this latter is stable by both inter-molecular and intramolecular hydrogen bonding. Up to 100 glucon chains are grouped in order to form a long thin crystallites which considered as an elementary fibrils, the cellulose source influence to the width of the crystallites [4]. They are organized in groups to form microfibrils that are between 5 and 80 nm in diameter and a few micro meters for length [5]. The crystal structure of this nanofibers make the plant stem more strength [6]. Chemical and mechanical treatment of the fibers cause a modification on the fiber surface and the cells, which influence the properties of the fibers in composites [7, 8]. The mallow fiber © The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 845–853, 2023. https://doi.org/10.1007/978-3-031-35248-5_74
846
Y. El Omari et al.
was used in lot of application like making cordage in clothes, carpets, paper money… [9]. But it has not heretofore been of great utility as reinforcing component in industrial composite products [10]. The Mallow fibers can be used as reinforcement in polymer matrix composites as shown by physical and mechanical properties [11]. The properties physical and mechanical of Mallow are presented in the Table 1 [12]. Table 1. Physical and mechanical properties of the mallow fibers [12]. Natural fiber
Diameter (µ m)
Length (mm)
Specific masse (kg/m3 )
Tensile strength (MPa)
Elongation at Modulus of break (%) elasticity (GPa)
Mallow
42.6
23.8
1374
160
5.2
17.4
Which prove that there is a possibility to use mallow fibers for polymer matrix composite reinforcement. Therefore, increased crystallinity of mallow fibers can improve mechanical properties of the polymer matrix composites by the mercerization treatment.
2 Experimental Processes Moroccan mallow stem fibers (MMFs) were collected from region of khouribga, in the winter weather Fig. 1 show Moroccan mallow plant and its extracted fibers. Table 1 indicates the comparison of various physicochemical constituents and mechanical properties of MMFs with other cellulosic fibers [12]. Extraction of Mallow Fibers The products are sampling in winter weather and they are save on roof of the House from winter to summer weather, the colour of plants change and there are transform to fibers. The plant of mallow was first immersed in water to allow microbial degradation for 14 days. The fibers were extracted from the mallow plant by retting method after degradation process. Then, the extracted fibers were dried in sunlight for a week to remove the moisture content. The fibers are cut into small pieces of length 4 mm and 3 mm in width, thus the samples have been prepared.
Extraction and Characterization of New Cellulosic Fibers
847
Fig. 1. (a) mallow plants (b) Extracted fiber collected from Moroccan mallow plant.
3 X-Rray Fluorescence 3.1 X-Ray Fluorescence (XRF) Analysis X-ray fluorescence (XRF) spectroscopy is one of the simplest and most widely used techniques for the non-destructive multi element analysis of materials. This technique has a remarkable progress in its applicability in material science. In conventional XRF, the element detection sensitivities are largely limited to the μg.g–1 (ppm) range. Mainly, the large spectral background produced by the Compton scattered X-rays from the specimen. The X-ray fluorescence spectrometer is capable to analysis elements in concentrations ranging from a high percentage up to the ppm level. In our case, the importance of the chemical element analysis is to determine the chemical composition of the cellulose fiber. The results show the existence of oligo elements and major elements. Oligo elements in ppm (parts per million) such as Cr2 O3 , Fe2 O3 , NiO …, and major elements such as 0.14% in SiO2 , 0.173% in P2 O5 , 0.278% in Cl , 0.489% in CaO , 0.102% in SO3 , the components which are in ppm are the origin of the clay and these components are absorbed by the plant as shown in Table 2.
848
Y. El Omari et al. Table 2. Chemical elements concentrations in mallow fiber samples
Major elements
Oligo elements
Samples
Mallow
Compound
Conc
Unit
CaO
0,489
%
SiO2
0,14
%
P2O5
0,173
%
SO3
0,102
%
Cl
0,278
%
Cr2O3
8,2
Ppm
V2O5
10,2
Ppm
MgO
456,7
Ppm
Al2O3
591,8
Ppm
K2O
788,3
Ppm
TiO2
290
Ppm
Fe2O3
119,7
Ppm
NiO
173,8
Ppm
ZnO
5,8
Ppm
Ag2O
201,9
Ppm
SnO2
0,5
Ppm
Re
0,4
Ppm
The XRF analysis was improved that the mallow’s fiber contain also the inorganic compound in considerable parties which based of composition of clay which describe the cristalinity of mallow fiber presented in this study [16, 17]. 3.2 X-Ray Diffraction (XRD) Analysis The mallow fibre sample was characterised by X-Ray Diffraction analysis (XRD), by this technic we can evaluated the crystallinity of the sample by wide-angle X-ray diffraction analysis using a Bruker D8 ADVANCE Powder XRD apparatus who is the most used. XRD allow the characterization of the mallow fibre microstructure. The use of the diffraction spectrum to go back to the crystallinity index of the analysed substrate can be done using different analytical methods [13]. We used the Eq. (1) to estimate the crystal fraction (CrI) from the height of two peaks, which corresponds to the crystalline and amorphous fractions baseline [13]. CrI (%) =
I002 − Iam × 100 I002
(1)
I002 : is the maximum intensity of the peak for 2θ = 22° which corresponds to the crystalline fraction.
Extraction and Characterization of New Cellulosic Fibers
849
Iam : is the minimum intensity of the peak for 2θ = 15° which corresponds to the amorphous fraction as shown in the Figs. 2 and 3. The Table 3 show the Crystallinity index CrI (%) for some fibers. Table 3. Crystallinity characteristics for some fibers [14] Fibers
CrI (%)
Untreated corn stem fibers
58.2
NaOH treated corn stem fibers
65.7
NaOH-silane treated corn stem fibers
64.4
Silane treated corn stem fibers
69.7
Fig. 2. Spectrum of X Ray diffraction patterns of Mallow fiber
CrI (%) =
7.2 − 4.1 × 100 7.2
CrI = 43% The Table 4 show that our fiber are crystalized in Orthorhombic system with cell Parameters a = 20.5008 A, b = 12.9599 A, c = 4.2672 A and cell volume V=1133.7345 A3 .
850
Y. El Omari et al.
Fig. 3. Profile matching by fullprof program for mallow plant
Table 4. Structural parameters mallow samples. Crystal System
Symmetry Cell Parameters (A) Group A B C
Orthorhombic P m m m
α (°) β (°) γ (°) Cell Volume (A3)
20.5008 12.9599 4.2672 90
90
90
1133.7345
3.3 Fourier Transformed Infra-red (FTIR) Spectral Analysis FT-IR spectra were used to examine the structure of Moroccan mallow fibers. For obtaining a spectra of each sample we used a Nicolet 560 spectrophotometer. The untreated and treated maize tassel fibers were grounded and mixed with KBr powders and the mixture was compressed into plates for FT-IR analysis. FTIR spectrometer was used to determine the presence of free functional groups on the MMF. Spectral outputs were obtained in the wavelength range of 4000–400 cm−1 using 32 scans and recorded in the transmittance mode as a function of wave number, the Figure 4 represent the spectrum of FT-IR of Mallow plant (Table 5). The FTIR spectra for mallow fiber is shown in Fig. 4. The characteristic peaks at approximately 3342 cm−1 correspond to the O − H bond stretching which indicate the presence of α-cellulose, the peak observed at 1730 cm−1 for C − O stretching of hemicellulose, the peak observed at 1637 cm−1 correspond to the C = C bond which indicate the presence of lignin, the peak observed at 1241,95 cm−1 correspond to C − O − C bond stretching from ether linkage of lignin, the peak observed at 1031,72 cm−1 correspond to the aromatic C −H stretching of lignin, the peak observed at 558,33 cm−1 correspond to aliphatic C − I stretching [15] (Table 6).
Extraction and Characterization of New Cellulosic Fibers
851
Table 5. Absorption bands for functional groups of cellulose and hemicellulose and lignin Fiber component
Wave number (cm−1 )
Functional group
Compounds
Cellulose
4,000–2,995 2,890 1,640 1,270–1,232 1,170–1,082 1,108
OH H–C–H Fiber –OH C–O–C C–O– C OH
Acid, methanol Alkyl, aliphatic Adsorbed water Aryl-alkyl ether Pyranose ring skeletal C- OH
Hemicellulose
4,000–2,995 2,890 1,765–1,715 1,108
OH H–C–H C=O OH
Acid, methanol Alkyl, aliphatic Ketone and carbonyl C–OH
Lignin
4,000–2,995 2,890 1,730–1,700 1,632 1,613, 1,450 1,430 1,270–1,232 1,215 1,108 700–900
OH H–C–H C=C C=C O–CH3 C–O–C C–O OH C–H
Acid, methanol Alkyl, aliphatic Aromatic Benzene stretching ring Aromtic skeletal mode Methoxyl– O–CH3 Aryl-alkyl ether Phenol C–OH Aromatic hydrogen
Fig. 4. Spectrum of FT-IR for mallow plant
In this Fig. 4, we are observed all peaks between 450 cm-1 and 4000 cm-1 ; these quantified the character of cellulose fibers of plants in this subject, therefor the absorbance is presented in the Fig. 5. The FTIR analysis show that our mallow fiber present all essential cellulosic vibration, which prove that our sample is in good cellulosic quality.
852
Y. El Omari et al.
Table 6. Peak positions and assignments of chemical groups in the untreated mallow fiber Samples
IR Band cm−1
Mallow fiber
3342,47
Transmittance % 98,09
Description O − H bond stretching
1730
103,5
C − O stretching
1637,62
103,21
C = C bond
1241,95
102,18
C − O − C bond stretching
1031,72
90,68
aromatic C − H stretching
558,33
97,20
aliphatic C − I stretching
Fig. 5. Spectra of absorbance for fibers of mallow plants
4 Conclusion This study show that our mallow fibers had other elements, lignin and hemicellulose, which decrease the crystallinity index. The results obtained from FT-IR analyses showed that our sample is good in cellulose content despite the presence of lignin and hemicellulose. This study advances the possibility of utilizing the mallow fibers, which is renewable, and biodegradable, as reinforcement in bio composites. In the further works, we will further investigate the effect of alkali-treatment with different percentage of NaOH on the morphological, chemical, and thermal properties of the Moroccan mallow fibers. Acknowledgements. Special thanks to dear Abderahim HASRI from Hassan I University of Settat, Morocco for XRD and XRF apparatus. Special thanks to dear Aziz Essoufi from Sultan Moulay Sliman university of Beni Mellal. Special thanks to dear Khalil Elhami from scientific institute of Rabat, Mohamed V university.
References 1. Gurunathan, T., Mohanty, S., Nayak, S.K.: A review of the recent developments in biocomposites based on natural fibres and their application perspectives. Compos. Part A (2015). http://dx.doi.org/https://doi.org/10.1016/j.compositesa.2015.06.007 2. Chand, N.: Tribology of Natural Fiber Polymer Composites. Nat. Fibers Their Compos. 1–58 (2008). https://doi.org/10.1533/9781845695057.1 3. Rongpipi, S., Ye, D., Gomez, E.D., Gomez, E.W.: Progress and opportunities in the characterization of cellulose – an important regulator of cell wall growth and mechanics. Front. Plant Sci. 9, 1894 (2019). https://doi.org/10.3389/fpls.2018.01894
Extraction and Characterization of New Cellulosic Fibers
853
4. Harris, D.M., et al.: Cellulose microfibril crystallinity is reduced by mutating C-terminal transmembrane region residues CESA1A903V and CESA3T942I of cellulose synthase. Proc. Natl. Acad. Sci. U.S.A. 109, 4098–4103 (2012). https://doi.org/10.1073/pnas.1200352109 5. Müller, M., Hori, R., Itoh, T., Sugiyama, J.: X-ray microbeam and electron diffraction experiments on developing xylem cell walls. Biomacromol 3, 182–186 (2002). https://doi.org/10. 1021/bm015605h 6. Agarwal, U.P.: 1064 nm FT-Raman spectroscopy for investigations of plant cell walls and other biomass materials. Front. Plant Sci. 5, 490 (2014). https://doi.org/10.3389/fpls.2014. 00490 7. Donaldson, L., Frankland, A.: Ultrastructure of iodine treated wood. Holzforschung 58, 219– 225 (2004). https://doi.org/10.1515/HF.2004.034 8. Astley, O.M., Chanliaud, E., Donald, A.M., Gidley, M.J.: Structure of Acetobacter cellulose composites in the hydrated state. Int. J. Biol. Macromol. 29, 193–202 (2001). https://doi.org/ 10.1016/S0141-8130(01)00167-2 9. Nieduszynski, I., Preston, R.D.: Crystallite size in natural cellulose. Nature 225, 273–274 (1970). https://doi.org/10.1038/225273a0 10. Atalla, R.H., Vanderhart, D.L.: Native cellulose: a composite of two distinct crystalline forms. Science 223, 283–285 (1984). https://doi.org/10.1126/science.223.4633.283 11. Ade, H., Hitchcock, A.P.: NEXAFS microscopy and resonant scattering: composition and orientation probed in real and reciprocal space. Polymer 49, 643–675 (2008). https://doi.org/ 10.1016/j.polymer.2007.10.030 12. Vignesh, V., et al.: Indian mallow fiber reinforced polyester composites: mechanical and thermal properties. J. Mater. Res. Technol. 11, 274–284 (2021). https://doi.org/10.1016/j. jmrt.2021.01.023 13. Elenga, R.G., Djemia, P., Tingaud, D., Chauveau, T., Maniongui, J.G., Dirras, G.: Effects of alkali treatment on the microstructure, composition, and properties of the Raffia textilis fiber. BioRes. 8(2), 2934–2949 (2013) 14. Liu, Y., Xie, J., Wu, N., Ma, Y., Menon, C., Tong, J.: Characterization of natural cellulose fiber from corn stalk waste subjected to different surface treatments. Cellulose 26(8), 4707–4719 (2019). https://doi.org/10.1007/s10570-019-02429-6 15. Nandiyanto, A.B.D., et al.: How to read and interpret FTIR spectroscope of organic material. Indones. J. Sci. Technol. 4(1), 97–118 (2019). https://doi.org/10.17509/ijost.v4i1.15806 16. Gourouza M., Zanguina A., Natatou I., Boos A., Rev.: CAMES – Sci. Struct. Mat. 1, 29 (2013) 17. Sadki, H., Ziat, K., Saidi, M.: J. Mater. Environ. Sci. 5(S1), 2061 (2014)
Author Index
A Aarrad, Manar 234 Aasfara, Jehane 359 Abbaoui, Wafae 91 Abdelbaki, Mohamed 785 Abdellaoui, Ayoub 294 Abdelmoti, Albatnan 599 Abdoun, Otman 576 Achoch, Mounia 391 Achoual, Khalid 700 Adib, Abdellah 469 Agliz, Driss 242 Ahami, A. 305 Ahlam, Hamim 735 Ahlem, Hamdache 693, 735 Ahmed, Bagayou 693 Ait Laasri, El Hassan 242 Aitzaouiat, Charafeddine 785 Aksasse, Brahim 58 Al Idrissi, Najib 349 Al idrissi, Najib 359 Al Idrissi, Najib 384 Albatnan, Abdelmoti 609 Alhyane, Rachid 757 alidrissi, Najib 805 Alilou, Hakim 637 Allal, Douira 599 Allali, Imane 805 Amal, El Khaddari 645 Amina, Ouazzani Touhami 599 Amkor, Ali 528 Ammari, Mohammed 272 Annouch, Anouar 327 Aqili, Nabil 587, 805 Arbelo, Manuel 174 Ardchir, Soufiane 284 Asri, Hiba 725 Assad, Noureddine 104 Atmani, Abderrahman 242 Ayad, Habib 469 Aymani, Ismail E L 479
Azeroual, Saadia 440 Azzaoui, F.-Z. 305 Azzouazi, Mohamed 284 Azzouzi, Hamid 46, 185 B Baali, El Houssain 566 Bahnasse, Ayoub 215 Ba-ichou, Ayoub 128 Barhamgi, Mahmoud 35 Batess, Rachid 566 Bekri, Ali 128 Belahbib, Nadia 700 Belhoussine Drissi, T. 426 Bellabdaoui, Adil 327 Bellarbi, Larbi 165 Belyamani, Lahcen 359, 805 Ben Allal, Laïla 272 Benaji, B. 405 Benali, Abdelhamid 416 Ben-Bouazza, Fatima-ezzahraa 440 Benhlima, Said 128 Benkirane, Rachid 479, 521, 609, 743 Benlghazi, Ahmad 416 Benouini, Rachid 451 Bensalah, Nouhaila 469 Benslimane, Djamal 35 Benslimane, Oumayma 462 Benyahia, H. 536 Benyahia, Hamid 623 Benyahya, Youssef 552 Berber, Fadoua 609 Berrahma, Chaimaa 384 Boualoulou, N. 426 Bouazza, Tbib 816, 829 Bouchar, Abdesslam 521 Bouchara, Aicha 272 Boudoudou, D. 536 Bougdira, Abdesselam 349 Bouhsine, Taha 174 Bouhssini, Ahmed 700
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 J. Kacprzyk et al. (Eds.): AI2SD 2022, LNNS 713, pp. 855–858, 2023. https://doi.org/10.1007/978-3-031-35248-5
856
Boukhris, Abdelouafi 725 Boulbaroud, S. 305 Bouni, Mohamed 284 Bouramtane, Khalil 250 Bourazza, Manal 623 Bouroumi, Abdelaziz 757 Bouskri, Ghizlane 566 Britel, Mohammed Reda 8 C Chaik, Asma 637, 683, 766 Chaima, Achagra 683 Cherkaoui, Souad 198 Cherrat, Loubna 1, 116, 637 Chetouani, Abdelaziz 416 Chetto, Ouiam 623 Chliyeh, Mohamed 743 D Dabghi, Aomar 700 Dahbi, Mohamed Reda 174 Dahmani, Jamila 700 Douira, A. 536 Douira, Allal 479, 521, 609, 623, 743 Drissi, E. 305 E Eddya, Mohammed 845 El Alaoui Moulay, Abdelaziz 743 El Bajta, Manal 773 El Barbri, Noureddine 528 El Beqqali, Omar 250 El Boustani, Abdelhakim 757 El Fellah, Younes 566 El Gabardi, Soumaya 743 El Ghoubali, Douae 349, 384 El Hajjaji, Souad 462 El Hannaoui, Oumayma 258 El Haoussi, Fatima 258 El Jgham, Bouchra 576 El Kaissoumi, Hanane 609 El Khatir, Haimoudi 576 El Kholfy, Saifeddine 521 El Marrakchi, S. 536 El Omari, Fatima 587 El Omari, Youssef 845 El Omary, Sara 490 El Ouafi, Zainab 349 El Ouazzani, Rajae 490
Author Index
El Yessefi, Abderrahim 1 Elfatouaki, Habiba 785 El-Hami, Khalil 845 Elouadi, Abdelmajid 84 Elouafi, Zainab 805 Elouark, M’hammed 479 Elouerghi, Achraf 165 Elyoussfialaoui, M.H 405 Errais, Reda 566 Es-sadek, Mohamed Zeriab 359 Es-Sadek, Mohamed Zeriab 587 Es-sadek, Mohamed Zeriab 805 Ezziyani, Mostafa 587, 805 Ezziyyani, Mohammed 637, 766 Ezziyyani, Mostafa 1, 75, 116, 349, 359, 637, 792 F Fail, Abderrahim 552 Farchi, Abdelmajid 136 Farhoune, Hassania 198 Farouk, Abdelhamid Ibn El 469 Fennan, Abdelhadi 19 Fettah, Amina 148 Feverati, Giovanni 391 G Gattal, Abdeljalil 148 Ghadi, Abderrahim 19 Ghalim, El Mehdi 84 Ghazal, Hassan 349, 359, 384, 587, 805 Ghoumari, Mohamed Yassine El 284 Gilabert, Catalina Egea 637 Gsim, Jamal 359 H Habib, Nihal 349 Hadni, Meryeme 773 Hamdache, Ahlem 637, 766 Hamdi, Salsabil 805 Hami, H. 305 Hamim, Ahlam 8 Hanafi, Hamza 318 Harama, Dounia 8 Haricha, Karim 215 Hassani, Badr Dine Rossi 318 Hilal, Mohamed 234 Houda, Zitan 368
Author Index
857
I Ichahane, Moulay Youssef 104 Idbraim, Soufiane 174 Idrissi Azami, Abdellah 349, 384 Imane, Mesbahi 735 Ismaili, Soumaya El 8 Issaoui, Yassine 215 Ivars-Palomares, Alberto 500 J Jamila, Dahmani 645 Jidane, Saïd 359, 805 Jilali, Antari 725 K Kadiri, Ahmed 46 Kamal, Saadi 645 Karima, Selmaoui 599 Karimi, Leila 8 Kassimi El Bakkali, Aziza 757 Kbir, M’hamed Aït 318 Kbiri, Zakaria 829, 845 Kerroum, B. 536 Khalid, Achoual 645 Khalid, Chafik 368 Kharmoum, Nassim 75, 91, 359, 587, 805 Kharraja, Said 250 Khelifi, Aymen 35 Khiat, Azeddine 215 Khomsi, Zakaryae 165 Khrouch, Sarah 1, 116 Krim, Mustapha 343 L Laaz, Naziha 673 Lahcen, Zidane 645 Lahrache, Souad 490 Laksiby, Ikram 637 Lamsellak, Oussama 416 Latif, Adnane 785 Lebbar, S. 405 Lemsayah, Mustapha 349 Lesieur, Claire 391 Lloret, Jaime 500 Lmoussaoui, Mohamed 242 M M’guil, Malika 8 Maazouzi, Soukaina
521
Machrouh, Youssef El 343 Manal, El Moudden 735 Maroua, Badri 683 Masse, Antoine 174 Maurady, Amal 8 Mauri, Jaime 637 Mbarki, Samir 673 Mchita, Mohamed 766 Mehdi, Imane 272 Menassel, Rafik 148 Meyma, Mint Mohamed 673 Mhamdi, Jamal 792 Millimono, Sory 587 Modafar, Cherkaoui E L 479 Mohamed, Faoussi 816, 829 Mohammed, Ezziyyani 683, 693, 735 Mokhtar, Naima Bel 8 Mouden, Najoua 479, 609 Mouhib, Ibtihal 773 Moukhtari, A. 405 Mouria, Afifa 743 Mousaid, Driss 416 Msairi, Soukaina 479, 521, 743 N Naanani, Hassan 773 Nadia, Belahbib 645 Najoua, Mouden 599 Naqi, Amine 440 Nejjari, Chakib 349, 384, 587 Nfaoui, El Habib 258 Nsiri, B. 405, 426 O Ouahdani, Houria El 700 ouahmane, Hassan 104 Ouajji, Hassan 215 Ouariachi, Ilham El 451 Ouassit, Youssef 284 Ouazzani Chahdi, Abdelatif 609 Ouazzani Touhami, Amina 521, 609, 743 Ouhda, Mohamed 58 Ourras, Samah 479 P Pirro, Stacy
384
Q Qouhafa, M. 405
858
R Rabie, Reda 462 Rachid, Benkirane 599 Rajaallah, El Mostafa 234, 343 Rami, Amal 359 Regragui, Hajar 185 Reklaoui, Kamal 185 Rémy, Florence 757 Remzan, Nihal 136 Retal, Sara 75, 91 Rhalem, Wajih 1, 75, 349, 359, 384, 587, 792, 805 Riffi, Jamal 250 S Saadi, Kamal 700 Sadik, Mohamed 552 Sahbani, Hajar 75 Said, Benchoucha 683 Sajadi Ansari, Fatemeh 35 Salamatian, Kave 391 Salim, Bounou 816, 829 Samih, Amina 19 Sara, Lahlou 515 Sebihi, Rajaa 440 Sedrati, Hayat 587, 805 Sefiani, Naoufal 185 Sefiani, Noufel 46 Sehli, Sofia 384 Sellal, Zineb 521 Selmaoui, Karima 479, 609, 743 Sendra, Sandra 500
Author Index
Soulaymani, A. 405 Soumia, Ziti 515 T Tahiry, Karim 136 Talha, A. 536 Talha, Abdelhak 623 Tbib, Bouazza 845 Touati, Iman 8 Touhami, A. Ouazzani 536 Touhami, Amina Ouazzani 479 Tsouli Fathi, Maroi 1, 116 Tsouli Fathi, Ramz 116 V Viciano-Tudela, Sandra 500 vuillon, Laurent 391 W Waga, Abderrahim Y Yassine, Diane
128
693
Z Zarghili, Arsalane 451 Zarouit, Yousra 58 Zenkouar, Khalid 451 Zidane, Ilyas 792 Zidane, Nordine 792 Ziti, Soumia 91