Artificial intelligence in medicine : 19th International Conference on Artificial Intelligence in Medicine, AIME 2021 : virtual event, June 15-18, 2021 : proceedings

This book constitutes the refereed proceedings of the 19th International Conference on Artificial Intelligence in Medicine, AIME 2021, held as a virtual event, in June 2021. The 28 full papers presented together with 30 short papers were selected from 138 submissions. The papers are grouped in topical sections on image analysis; predictive modelling; temporal data analysis; unsupervised learning; planning and decision support; deep learning; natural language processing; and knowledge representation and rule mining. Preface Organization Transforming Healthcare with Artificial Intelligence – Lessons from Ophthalmology (Abstract of Invited Talk) Contents Invited Talk The Myth of Complete AI-Fairness References Image Analysis A Petri Dish for Histopathology Image Analysis 1 Introduction 2 Background 3 MHIST Dataset 3.1 Colorectal Polyp Classification Task 3.2 Data Collection 3.3 Data Annotation 4 Example Use Cases 4.1 Experimental Setup 4.2 Network Depth 4.3 Transfer Learning 4.4 High-Disagreement Training Examples 5 Related Work 6 Discussion References fMRI Multiple Missing Values Imputation Regularized by a Recurrent Denoiser 1 Introduction 2 Problem Setting 3 Proposed Approach 3.1 Spatial Imputation 3.2 Time Dimension Regularization 3.3 Validation and Hyperparameters 4 Related Work 5 Experimental Setting 5.1 Baselines 5.2 Datasets 5.3 Random Value and Region Removal 5.4 Evaluation Metrics 6 Experiment Results 7 Concluding Remarks References Bayesian Deep Active Learning for Medical Image Analysis 1 Motivation 2 Proposed Method 2.1 Estimating Confidence in Image Classification 2.2 Calibrated Uncertainty in Semantic Image Segmentation 3 Application of Active Learning for Medical Image Analysis 3.1 Image Classification 3.2 Semantic Segmentation 4 Conclusion and Future Research References A Topological Data Analysis Mapper of the Ovarian Folliculogenesis Based on MALDI Mass Spectrometry Imaging Proteomics 1 Introduction 2 Exploratory Analysis and Dimensionality Reduction 3 TDA Mapper Approach 4 Conclusions References Predictive Modelling Predicting Kidney Transplant Survival Using Multiple Feature Representations for HLAs 1 Introduction 2 Background and Motivation 3 Data Description 4 Methods and Technical Solutions 4.1 Feature Representations 4.2 Survival Analysis Algorithms 5 Empirical Evaluation 5.1 Evaluation Metric 5.2 Effects of Feature Representations 6 Related Work 7 Significance and Impact References Sum-Product Networks for Early Outbreak Detection of Emerging Diseases 1 Introduction 2 Non-specific Syndromic Surveillance 2.1 Problem Definition 2.2 Creation of Structured Data 2.3 Related Work 3 Sum-Product Networks for Syndromic Surveillance 3.1 Inference of p-values in Sum-Product Networks 3.2 Application to Non-specific Syndromic Surveillance 3.3 Handling of Higher Order Syndromes 4 Experiments and Results 4.1 Evaluation Setup 4.2 Results 5 Conclusion References Catching Patient's Attention at the Right Time to Help Them Undergo Behavioural Change: Stress Classification Experiment from Blood Volume Pulse 1 Introduction 2 Related Work 3 Methods 3.1 Dataset 3.2 Stress Detector 3.3 Experiments 4 Results 5 Discussion References Primary Care Datasets for Early Lung Cancer Detection: An AI Led Approach 1 Introduction 2 Related Work 3 Dataset Description 3.1 NPS MedicineInsight 3.2 Cancer Patient’s Analysis 4 Features Design and Methods Selection 4.1 Uncertainty Based Features Design 4.2 Soft Out of Range Results 4.3 Additional Features 5 Model Selection, Experiments and Results 5.1 Model Selection 5.2 Experiments and Results 6 Conclusion References Addressing Extreme Imbalance for Detecting Medications Mentioned in Twitter User Timelines 1 Introduction 2 Methods 3 Results and Discussion 3.1 Intrinsic Evaluation Results 3.2 Extrinsic Evaluation Results 4 Conclusion References ICU Days-to-Discharge Analysis with Machine Learning Technology 1 Introduction 2 Methods and Results 2.1 The Dataset 2.2 Measuring Heterogeneity 2.3 Identification of DTD Biomarkers 2.4 Phenotype Extraction 2.5 Building DTD Predictive Models 3 Discussion 4 Conclusion References Transformers for Multi-label Classification of Medical Text: An Empirical Comparison 1 Introduction 2 Related Work 3 Data 4 Neural Network Algorithms 4.1 Transformers 4.2 Traditional Neural Networks 5 Experiments 6 Results 7 Conclusions References Semantic Web Framework to Computerize Staged Reflex Testing Protocols to Mitigate Underutilization of Pathology Tests for Diagnosing Pituitary Disorders 1 Introduction 2 Reflex and Reflective Testing Protocols for Pituitary Disorder 3 Semantic Framework for Computerizing Reflex Protocols 4 Implementation of the Pituitary Disorder System 5 Evaluation of Accuracy and Cost Effectiveness 6 Conclusions and Future Work References Using Distribution Divergence to Predict Changes in the Performance of Clinical Predictive Models 1 Introduction 2 Related Work 2.1 Divergence Metrics to Predict Changes in Model Performance 2.2 Distribution Divergence in Healthcare Data 3 Divergence Metrics 3.1 Wasserstein Distance (WD) 3.2 Maximum Mean Discrepancy (MMD) 3.3 MMD p-value 4 Methods 4.1 Dataset 4.2 Experimental Setup 4.3 Dataset Preparation 4.4 Evaluation 5 Results 6 Discussion and Conclusion References Analysis of Health Screening Records Using Interpretations of Predictive Models 1 Introduction 2 Related Work 3 Method 4 Evaluation 5 Conclusion References Seasonality in Infection Predictions Using Interpretable Models for High Dimensional Imbalanced Datasets 1 Introduction 2 Methods 3 Experiment 4 Results 5 Discussion and Conclusions References Monitoring Quality of Life Indicators at Home from Sparse, and Low-Cost Sensor Data 1 Introduction 2 Methods 2.1 Data Collection 2.2 Model Development and Evaluation 3 Results 3.1 Model Accuracy 3.2 Activity Accuracy 4 Discussion and Conclusions References Detection of Parkinson's Disease Early Progressors Using Routine Clinical Predictors 1 Motivation 2 Patients 3 Predictors 4 Methods 5 Results and Discussion References Detecting Mild Cognitive Impairment Using Smooth Pursuit and a Modified Corsi Task 1 Introduction 2 Patients and Methods 2.1 Smooth Pursuit Test 2.2 Modified Corsi Block-Tapping Test 3 Results 4 Discussion and Conclusions References Temporal Data Analysis Neural Clinical Event Sequence Prediction Through Personalized Online Adaptive Learning 1 Introduction 2 Related Work 3 Methodology 3.1 Neural Autoregressive Event Sequence Prediction 3.2 Online Adaptation of Model Parameters 3.3 Adaptation by Model Switching 4 Experimental Evaluation 4.1 Experiment Setup 4.2 Results on Online Adaptation vs. Population Model 4.3 Results on Adaptation on Partial Components 4.4 Results for Online Switching-Based Adaptation 5 Conclusion References Using Event-Based Web-Scraping Methods and Bidirectional Transformers to Characterize COVID-19 Outbreaks in Food Production and Retail Settings 1 Introduction 2 Background 2.1 COVID-19 and Food Establishments 2.2 Bidirectional Transformers 3 Methods 3.1 NLP-Based Web-Scraping 3.2 Data Validation 3.3 Applying RoBERTa to Article Classification 4 Results 4.1 Web-Scraping, Manual Data Validation, and Visualization 4.2 RoBERTa Model Evaluation 5 Discussion 5.1 RoBERTa Model and NAICS Codes 5.2 COVID-19 Surveillance 6 Conclusion 7 Disclaimer References Deep Kernel Learning for Mortality Prediction in the Face of Temporal Shift 1 Introduction 2 Deep Kernel Learning 3 Experiments 3.1 Temporal Shift: Strategy and Results 3.2 Experiment 2: Internal Validation 4 Related Work 5 Conclusions and Future Work References Model Evaluation Approaches for Human Activity Recognition from Time-Series Data 1 Introduction 2 Materials and Methods 3 Results and Discussion 4 Conclusion References Unsupervised Learning Unsupervised Learning to Subphenotype Heart Failure Patients from Electronic Health Records 1 Introduction 2 Methods 2.1 Data Source and Phenotyping Algorithm 2.2 Features and Patient Embedding 2.3 Clustering 2.4 Evaluation 3 Results 3.1 Results from the Aggregated Data Strategy 3.2 Results from the Sequence Data Strategy 4 Discussion 4.1 Technical Significance 4.2 Clinical Significance 4.3 Limitations and Future Work 5 Conclusion References Stratification of Parkinson's Disease Patients via Multi-view Clustering 1 Introduction 2 Data 3 Multi-view Clustering Methodology 4 Results 5 Conclusion References Disentangled Hyperspherical Clustering for Sepsis Phenotyping 1 Introduction 2 Method 2.1 Soft F-statistic 2.2 Measurements 3 Experiments 4 Results and Discussions References Phenotypes for Resistant Bacteria Infections Using an Efficient Subgroup Discovery Algorithm 1 Introduction 2 Methods 2.1 Algorithm Proposal 2.2 Subgroup Discovery Process for Phenotypes 3 Experiments and Dataset 4 Results and Discussion 5 Conclusions and Future Work References Predicting Drug-Drug Interactions from Heterogeneous Data: An Embedding Approach 1 Introduction 2 Embeddings Using Heterogeneous Data Sources 3 Empirical Evaluation References Detection of Junctional Ectopic Tachycardia by Central Venous Pressure 1 Background and Introduction 2 Methods 3 Experiment and Results 4 Conclusion References Planning and Decision Support A Cautionary Tale on Using Covid-19 Data for Machine Learning 1 Introduction 2 Methods 2.1 Work Processes 2.2 Registry Forms 3 Results 3.1 Work Processes 3.2 Registry Forms 4 Discussion 4.1 Workflow Processes 4.2 Registry Forms References MitPlan 2.0: Enhanced Support for Multi-morbid Patient Management Using Planning 1 Introduction 2 MitPlan 2.0 3 Illustrative Example 3.1 Cost Optimization 3.2 Revision Application 3.3 Clinical Illustrative Example 4 Discussion and Future Work References Explanations in Digital Health: The Case of Supporting People Lifestyles 1 Introduction 2 Related Work 3 The Reasoning-Based Explainable Approach 4 Evaluation 5 Conclusion References Predicting Medical Interventions from Vital Parameters: Towards a Decision Support System for Remote Patient Monitoring 1 Introduction 2 Materials and Methods 3 Results 4 Related Work 5 Conclusion References CAncer PAtients Better Life Experience (CAPABLE) First Proof-of-Concept Demonstration 1 Introduction 2 Methods 2.1 Consortium and Expertise 2.2 Iterative Development of the First POC and Its Components 2.3 Data Model and FAIR Principles 2.4 AI 3 Results 4 Conclusion References Deep Learning Sensitivity and Specificity Evaluation of Deep Learning Models for Detection of Pneumoperitoneum on Chest Radiographs 1 Introduction 2 Materials and Methods 2.1 Pneumoperitoneum Dataset and Expert Annotations 2.2 Training, Validation and Test Split 2.3 Deep Learning Methods 3 Results 3.1 Model Visualization and Error Analysis 3.2 Discussion 4 Conclusion References An Application of Recurrent Neural Networks for Estimating the Prognosis of COVID-19 Patients in Northern Italy 1 Introduction 2 Related Work 3 Available Data 4 Description of the Datasets 5 Recurrent Neural Network Model 5.1 Gated Recurrent Units and Attention Mechanism 5.2 Loss Function and Tuning of the Hyperparameters 6 Experimental Results 7 Conclusions and Future Work References Recurrent Neural Network to Predict Renal Function Impairment in Diabetic Patients via Longitudinal Routine Check-up Data 1 Introduction 2 Prediction Target and Study Population 2.1 Prediction Target: Impaired Kidney Function on the KDOQI Scale 2.2 Study Population and Dataset Split 3 Methods 3.1 Input Data Preparation 3.2 Output Coding 3.3 Model Architecture and Development 3.4 Performance Evaluation and Secondary Analyses 4 Results 5 Discussion and Conclusions References Counterfactual Explanations for Survival Prediction of Cardiovascular ICU Patients 1 Introduction 2 Counterfactual Explanations for Medical Sequences 2.1 Problem Formulation 2.2 Style-Transfer Counterfactual Explanations 2.3 Nearest Neighbour Counterfactual Explanations 3 Experiments 3.1 Experimental Setup 3.2 Empirical Investigation 4 Conclusions References Improving the Performance of Melanoma Detection in Dermoscopy Images Using Deep CNN Features 1 Introduction 2 Proposed Methodology 2.1 Boundary Localization 2.2 Feature Extraction 2.3 Classification 3 Experimental Setting and Results 3.1 Experimental Setting 3.2 Experiments with Proposed Method 3.3 Comparison with State-of-the-Art Methods 4 Conclusion References Mobile Aided System of Deep-Learning Based Cataract Grading from Fundus Images 1 Introduction 2 Novel Method for Cataract Grading 2.1 Preprocessing and Data Augmentation 2.2 Fine Tuning and Transfer Learning 2.3 Classification for Cataract Grading with Random Forest 3 Experimental Results 3.1 Dataset and Evaluation Metrics 3.2 Cataract Grading Performance 4 Mobile-Aided-Grading System for Cataract Disease 5 Conclusion References Uncertainty Estimation in SARS-CoV-2 B-Cell Epitope Prediction for Vaccine Development 1 Introduction 2 Cost-Sensitive Calibrated Uncertainty 3 Experiment 4 Experimental Results 4.1 Model Performance 4.2 Distribution of Uncertainty Estimates 4.3 The Contribution of Uncertainty Thresholds in Predictive Probabilities 5 Conclusion and Future Work References Attention-Based Explanation in a Deep Learning Model For Classifying Radiology Reports 1 Introduction and Background 2 Classification of Radiology Reports 3 Analysis of the Attention Mechanism 4 Conclusions References Evaluation of Encoder-Decoder Architectures for Automatic Skin Lesion Segmentation 1 Introduction 2 Proposed Methodology 3 Results 4 Conclusion References A Novel Deep Learning Model for COVID-19 Detection from Combined Heterogeneous X-ray and CT Chest Images 1 Introduction 2 DarkCovidNet-NRC DL Architecture 3 Experiments and Results 3.1 Dataset 3.2 Experiment Settings 3.3 Results and Discussion 4 Conclusion and Perspectives References An Experiment Environment for Definition, Training and Evaluation of Electrocardiogram-Based AI Models 1 Introduction 2 Components 2.1 Configuration 2.2 Data Sources and Snapshots 2.3 Data Preprocessing and Splitting 2.4 Experiment Conduction 2.5 Experiment Evaluation 3 Proof of Concept 4 Conclusion and Future Work References Enhancing the Value of Counterfactual Explanations for Deep Learning 1 Introduction 2 Background 3 Method 4 Clinical Example 5 Conclusion References Natural Language Processing A Multi-instance Multi-label Weakly Supervised Approach for Dealing with Emerging MeSH Descriptors 1 Introduction 2 Related Work 3 Obtaining Weak Supervision with WeakMeSH 3.1 Candidate Labels Generation 3.2 Multi-instance Semantic Similarity 3.3 Multi-instance Multi-label Learning from Weak Supervision 4 A Real-World Benchmark for Weakly Supervised Learning 5 Experimental Setup and Results 6 Conclusions and Next Steps References Demographic Aware Probabilistic Medical Knowledge Graph Embeddings of Electronic Medical Records 1 Introduction 2 Related Work 3 DARLING 3.1 Probabilistic Medical Knowledge Graph with Demographics 3.2 Demographic-Guided Translation 3.3 Optimization Through Probability Score 4 Experiments 4.1 Datasets and Medical Knowledge Graph Construction 4.2 Models for Comparison 4.3 Results 4.4 Demographic and Probability Score Sensitivity 5 Conclusions References Modeling and Representation by Graphs of the Reasoning of an Emergency Doctor: Symptom Checker MedVir 1 Introduction 2 Reasoning Model of an Emergency Physician 2.1 The Neuronal Entity Concept 2.2 Questions Priorization Algorithm [QPA] 2.3 Scenario Management 3 Visualisation by Graphs 3.1 Pyelonephritis Graph 4 Conclusion and Perspectives References Effect of Depth Order on Iterative Nested Named Entity Recognition Models 1 Introduction 2 Method 2.1 Model 2.2 Greedy Order Training 2.3 Model Parameters 3 Experiments and Discussions 3.1 Datasets 3.2 Baselines 3.3 Results 4 Conclusion References The Effectiveness of Phrase Skip-Gram in Primary Care NLP for the Prediction of Lung Cancer 1 Introduction 2 Materials and Methods 3 Results 4 Conclusions References Customized Neural Predictive Medical Text: A Use-Case on Caregivers 1 Introduction 2 Empirical Evaluation 3 Conclusion References Outlier Detection for GP Referrals in Otorhinolaryngology 1 Introduction 2 Background and Related Works 3 Design Overview 3.1 Uniform Manifold Approximation and Projection (UMAP) and Local Outlier Factor (LOF) 4 Data Preparation and Preprocessing 5 Experiment Results 6 Conclusion References The Champollion Project: Automatic Structuration of Clinical Features from Medical Records 1 Introduction 2 Methods 2.1 Project Definition 2.2 Sentences Annotation – Training Datasets 2.3 Training Machine Learning Classifiers 2.4 Automatic Structuration of Data 2.5 External Validation 3 Results and Conclusions References Knowledge Representation and Rule Mining Modelling and Assessment of One-Drug Dose Titration 1 Introduction 2 A 3-Step Modelling of Drug Dose Titration Procedures 2.1 Basic Components 2.2 Basic Procedures 2.3 Meta-models 3 Clinical Practice Supervision with DT Models 4 Conclusions References TransICD: Transformer Based Code-Wise Attention Model for Explainable ICD Coding 1 Introduction 2 Related Works 3 Dataset 4 Methods 4.1 Problem Formulation 4.2 Transformer Based Label Attention Model 5 Training Details 6 Evaluation 6.1 Results 6.2 Distribution of Scores 6.3 Visualization 7 Conclusion References Improving Prediction of Low-Prior Clinical Events with Simultaneous General Patient-State Representation Learning 1 Introduction 2 Related Work 3 Methodology 3.1 Model Definition 3.2 Optimizing Weighted Simultaneous Learning Loss 4 Experiments 4.1 Simultaneous Model Architectures 4.2 Baseline Model Architectures 4.3 Low-Prior Targets 4.4 Inputs and GPSR Tasks 4.5 Model Training and Selection 4.6 Weighted Loss Selection 5 Results and Discussion 5.1 Predictive Performance 5.2 Reduction of Prior Likelihood 5.3 Reduction of Sample Size 6 Conclusion References Identifying Symptom Clusters Through Association Rule Mining 1 Introduction 2 Modeling Symptom Clusters with ARM 3 Experimental Results 4 Conclusion References A Probabilistic Approach to Extract Qualitative Knowledge for Early Prediction of Gestational Diabetes 1 Introduction 2 Extracting Qualitative Influences 2.1 Proposed Approach 3 Learning Qualitative Influences for GDM Modeling References Author Index