Explainable AI in Healthcare and Medicine: Building a Culture of Transparency and Accountability

Preface
Contents
List of Contributors
Abbreviations
Explainability and Interpretability: Keys to Deep Medicine
1 Introduction
2 Explainability
3 Interoperability
4 Artificial Intelligence Tools and Methods
References
Fast Similar Patient Retrieval from Large Scale Healthcare Data: A Deep Learning-Based Binary Hashing Approach
1 Introduction
2 Related Work
3 Methods
3.1 The Artificial Neural Network
3.2 Similar Patient Retrieval
3.3 Evaluation
4 Experiments
4.1 Similar Patient Retrieval Efficiency
4.2 Similar Patient Retrieval Accuracy
4.3 Interpretation of the Results
5 Conclusions
References
A Kernel to Exploit Informative Missingness in Multivariate Time Series from EHRs
1 Introduction
2 Time Series Cluster Kernel to Exploit Informative Missingness
3 Experiments
4 Results and Discussion
References
Machine Learning Discrimination of Parkinson's Disease Stages from Walker-Mounted Sensors Data
1 Introduction
2 Our Approach: Machine Learning Enabled PD Stage Discrimination from Low-Cost Walker Mounted Sensor
3 Experimental Results
4 Discussion
References
Personalized Dual-Hormone Control for Type 1 Diabetes Using Deep Reinforcement Learning
1 Introduction
2 Related Work and Preliminaries
3 Methods
3.1 Generalized DQN Training
3.2 Personalized DQN Training
4 Experiment Results
5 Conclusion
References
A Generalizable Method for Automated Quality Control of Functional Neuroimaging Datasets
1 Introduction
2 Background
3 Data Sources
3.1 EMBARC
3.2 CNP
4 Methods
4.1 MRI Preprocessing Pipeline
4.2 Feature Sets
4.3 Feature Selection
4.4 Expert Manual QC Labels
5 Results
5.1 Predictive Models
5.2 Within Dataset Cross-Validation
5.3 Unseen Study Dataset as Test Set
6 Conclusion
References
Uncertainty Characterization for Predictive Analytics with Clinical Time Series Data
1 Introduction
2 Methods
3 Experiments and Results
3.1 Performance Results
3.2 Epistemic Uncertainty
3.3 Aleatoric Uncertainty
3.4 Implications for Prediction Performance
4 Conclusions
References
A Dynamic Deep Neural Network for Multimodal Clinical Data Analysis
1 Introduction
2 Related Work
3 Methods
3.1 Disease Progression Prediction
3.2 Sample Generation from Clinical Data
3.3 AdaptiveNet
4 Experimental Setup
4.1 Data Set
4.2 Disease Progression Prediction in RA
4.3 Baselines
4.4 Training
5 Results
6 Conclusion
References
DeStress: Deep Learning for Unsupervised Identification of Mental Stress in Firefighters from Heart-Rate Variability (HRV) Data
1 Introduction
2 Results and Discussion
2.1 Classical Feature Engineering and K-Means
2.2 Convolutional and LSTM AutoEncoders
3 Data Acquisition
4 Methods
4.1 K-Means Clustering
4.2 AutoEncoders
4.3 DBSCAN Clustering
4.4 Training and Evaluation
5 Conclusions
References
A Deep Learning Approach for Classifying Nonalcoholic Steatohepatitis Patients from Nonalcoholic Fatty Liver Disease Patients Using Electronic Medical Records
1 Introduction
2 Data and Method
3 Proposed Model Framework
4 Experiment
5 Conclusion
References
Visualization of Deep Models on Nursing Notes and Physiological Data for Predicting Health Outcomes Through Temporal Sliding Windows
1 Introduction
2 Related Work
2.1 Clinical Early Warning Systems
2.2 Neural Network Approaches
2.3 Interpreting Prediction Outcomes
3 Methods
3.1 Data
3.2 Models
4 Results
4.1 How Do Our Models Trained on Tabular Data Compare to Models Trained on Nurse Notes Data?
4.2 How Early Can Our Deep Models Make the Correct Prediction?
4.3 What Information Are the Deep Models Using to Make the Predictions?
4.4 What Are the Top Contributing Factors of Our Deep Models?
5 Conclusion
References
Constructing Artificial Data for Fine-Tuning for Low-Resource Biomedical Text Tagging with Applications in PICO Annotation
1 Introduction
2 Related Works
2.1 Multi-label Text Classification
2.2 Biomedical Text Annotation
2.3 Transfer Learning for Text Classification
3 Method
3.1 Model
3.2 Source Task: MeSH Annotation
3.3 Target Task: PICO Annotation
4 Experimental Setup
4.1 Dataset
4.2 Baselines
4.3 Evaluation Details
4.4 Metrics
5 Results
6 Conclusion
References
Character-Level Japanese Text Generation with Attention Mechanism for Chest Radiography Diagnosis
1 Introduction
2 Materials and Methods
2.1 Datasets
2.2 Overview of Methods
2.3 Learning Condition
2.4 Evaluation Metrics
3 Results and Discussion
4 Conclusion
References
Extracting Structured Data from Physician-Patient Conversations by Predicting Noteworthy Utterances
1 Introduction
2 Related Work
3 Dataset
3.1 Relevant Diagnosis Prediction
3.2 Review of Systems (RoS) Abnormality Prediction
4 Methods
4.1 Input-Agnostic Baseline
4.2 Medical-Entity-Matching Baseline
4.3 Learning Based Methods
4.4 Hybrid Models
5 Results and Discussion
5.1 Metrics
5.2 Results
5.3 Experimental Details
6 Conclusion and Future Work
References
A Multi-talent Healthcare AI Bot Platform
1 Introduction
2 System Architecture
2.1 Modular Design
2.2 The Pipeline
2.3 App Platform
2.4 Knowledge Base
2.5 Natural Language Understanding
3 Evaluation
3.1 Intent Classification
3.2 Entity Extraction
4 Conclusion
References
Natural vs. Artificially Sweet Tweets: Characterizing Discussions of Non-nutritive Sweeteners on Twitter
1 Introduction
1.1 Background: New Sweeteners
1.2 Stevia as a Natural Alternative
1.3 The Role of Social Media
2 Data
3 Topic Analysis
4 Temporal Patterns
References
On-line (TweetNet) and Off-line (EpiNet): The Distinctive Structures of the Infectious
1 Introduction
2 Data
3 Methods
4 Results
5 Findings and Discussions
References
Medication Regimen Extraction from Medical Conversations
1 Introduction
2 Data
3 Approach
3.1 Pointer-Generator Network (PGNet)
3.2 QA PGNet
3.3 Multi-decoder (MD) QA PGNet
4 Experiments
4.1 Data Processing
4.2 Metrics
4.3 Model Variations
5 Results and Discussion
5.1 Difference in Networks and Approaches
5.2 Breakdown of Performance
5.3 Training Dataset Size
5.4 Evaluating on ASR Transcripts
6 Conclusion
References
Quantitative Evaluation of Emergency Medicine Resident's Non-technical Skills Based on Trajectory and Conversation Analysis
1 Introduction
2 Non-technical Skills Evaluation Method for Emergency Medicine Residents
2.1 Data Acquisition Process
2.2 Clustering Method for Workflow Event Database Creation
2.3 Scoring Method Based on Workflow Event Database
3 Experiment and Discussion
3.1 Experimental Design
3.2 Results of Workflow Event Database Creation and Scoring Non-technical Skills
3.3 Discussion
4 Conclusion
References
Implementation of a Personal Health Library (PHL) to Support Chronic Disease Self-Management
1 Introduction
2 Method
3 The PHL in Action
4 Conclusion
References
KELSA: A Knowledge-Enriched Local Sequence Alignment Algorithm for Comparing Patient Medical Records
1 Introduction
2 Methods
2.1 Knowledge-Enriched Local Sequence Alignment Algorithm (KELSA)
2.2 Initialization of the Accumulated Score Matrix
2.3 Iterative Scoring Process to Fill the Accumulated Score Matrix
2.4 Traceback to Identify a Local Alignment
3 Evaluation
3.1 Evaluation Design
3.2 Real-World EHR Database for Evaluation
3.3 Synthesis of Patient Medical Records
3.4 Evaluation Metrics
4 Results and Discussion
5 Conclusion
References
Multi-Level Embedding with Topic Modeling on Electronic Health Records for Predicting Depression
1 Introduction
2 Methodology
3 Results
4 Conclusion
References
Faster Clinical Time Series Classification with Filter Based Feature Engineering Tree Boosting Methods
1 Introduction
2 Related Work
3 MIMIC-III Benchmark Task
3.1 In Hospital Mortality Prediction
3.2 25 Acute Care Phenotype Classification
4 Methods
4.1 RNN on Raw Data
4.2 Filter Based Feature Engineering (FBFE)
4.3 Tree Boosting Methods with FBFE
4.4 Two-Phase Auto Hyperparameter Optimization
5 Experiments
5.1 In Hospital Mortality Prediction
5.2 25 Phenotype Classification
6 Conclusion
References
Explaining Models by Propagating Shapley Values of Local Components
1 Introduction
2 Approach
2.1 Propagating SHAP Values
2.2 SHAP Values with a Background Distribution
3 Experiments
4 Conclusion
References
Controlling for Confounding Variables: Accounting for Dataset Bias in Classifying Patient-Provider Interactions
1 Introduction
2 Previous Work
3 Task: Classifying Goals-of-Care
4 Dataset
5 Feature Extraction and Backdoor Adjustment
6 Results and Analysis
7 Conclusion
References
Learning Representations to Augment Statistical Analysis of Drug Effects on Nerve Tissues
1 Introduction
2 Network Architecture
3 Experiments
3.1 Deep Neural Networks
3.2 Bioinformatic Statistical Analysis
4 Conclusion
References
Automatic Segregation and Classification of Inclusion and Exclusion Criteria of Clinical Trials to Improve Patient Eligibility Matching
1 Introduction
2 Attention Aware CNN-BiLSTM Model for Criteria Classification
3 Experiment and Results
4 Conclusion
References
Tell Me About Your Day: Designing a Conversational Agent for Time and Stress Management
1 Introduction
2 Background and Related Work
2.1 Stress and Health
2.2 Time Management and Coping
2.3 Conversational AI and Support
3 Proposed Approach
3.1 System Components
3.2 Implementation
4 Conclusions and Future Work
4.1 Designing Conversational Interventions for Stress
4.2 Improved Handling of Complex Temporal Expressions
4.3 Enhancing User Modeling with Self Disclosure
References
Accelerating Psychometric Screening Tests with Prior Information
1 Introduction
2 Bayesian Active Differential Inference
2.1 Bayesian Active Differential Inference for Gaussian Processes
3 Discussion
3.1 Related Work
4 Conclusion
References
Can an Algorithm Be My Healthcare Proxy?
1 Introduction
1.1 A Very Brief History of Advance Care Planning
2 Looking Forward: An ACP Decision Aid
2.1 Applications
3 Discussion
References
Predicting Mortality in Liver Transplant Candidates
1 Introduction
2 Related Work
3 Dataset
4 Predicting Mortality
5 Discussion
References
Towards Automated Performance Status Assessment: Temporal Alignment of Motion Skeleton Time Series
1 Background and Motivation
2 Related Works
3 Methods
3.1 Multivariate Time Series and Temporal Alignment
3.2 Dynamic Time Warping
3.3 Handling Task Left-Right Invariance
3.4 Handling Vertical-Horizontal Relative Importance
4 Experiments
4.1 Dataset
4.2 Evaluation Metrics
5 Results
6 Discussion
References