Intelligent Systems in Medicine and Health: The Role of AI

Foreword
Preface
The State of AI in Medicine
Introducing Intelligent Systems in Medicine and Health: The Role of AI
Structure and Content
Guide to Use of This Book
Acknowledgments
Contents
Contributors
Part I: Introduction
Chapter 1: Introducing AI in Medicine
The Rise of AIM
Knowledge-Based Systems
Neural Networks and Deep Learning
Machine Learning and Medical Practice
The Scope of AIM
From Accurate Predictions to Clinically Useful AIM
The Cognitive Informatics Perspective
Why CI?
The Complementarity of Human and Machine Intelligence
Mediating Safe and Effective Human Use of AI-Based Tools
Concluding Remarks
References
Chapter 2: AI in Medicine: Some Pertinent History
Introduction
Artificial Intelligence: The Early Years
Modern History of AI
AI Meets Medicine and Biology: The 1960s and 1970s
Emergence of AIM Research at Stanford University
Three Influential AIM Research Projects from the 1970s
INTERNIST-1/QMR
CASNET
MYCIN
Cognitive Science and AIM
Reflecting on the 1970s
Evolution of AIM During the 1980s and 1990s
AI Spring and Summer Give Way to AI Winter
AIM Deals with the Tumult of the 80s and 90s
The Last 20 Years: Both AI and AIM Come of Age
References
Chapter 3: Data and Computation: A Contemporary Landscape
Understanding the World Through Data and Computation
Types of Data Relevant to Biomedicine
Knowing Through Computation
Motivational Example
Computational Landscape
Knowledge Representation
Machine Learning
Data Integration to Better Understand Medicine: Multimodal, Multi-Scale Models
Distributed/Networked Computing
Data Federation Models
Interoperability
Computational Aspects of Privacy
Trends and Future Challenges
Ground Truth
Open Science and Mechanisms for Open data
Data as a Public Good
References
Part II: Approaches
Chapter 4: Knowledge-Based Systems in Medicine
What Is a Knowledge-Based System?
How Is Knowledge Represented in a Computer?
Rules: Inference Steps
Patterns: Matching
Probabilistic Models
Naive Bayes
Bayesian Networks
Decision Analysis and Influence Diagrams
Causal Mechanisms: How Things Work
How Is Knowledge Acquired?
Ontologies and Their Tools
Knowledge in the Era of Machine Learning
Incorporating Knowledge into Machine Learning Models
Graph-Based Models
Graph Representation Learning
Biomedical Applications of Graph Machine Learning
Text-Based Models
Leveraging Expert Systems to Train Models
Looking Forward
References
Chapter 5: Clinical Cognition and AI: From Emulation to Symbiosis
Augmenting Human Expertise: Motivating Examples
Cognitive Science and Clinical Cognition
Symbolic Representations of Clinical Information
Clinical Text Understanding
Clinical Cognition, Reasoning and the Evolution of AI
Bridging Cognition to Medical Reasoning
Models of Medical Reasoning
Knowledge Organization, Expert Perception and Memory
Understanding Clinical Practice for AI Systems
The Role of Distributed Cognition
AI, Machine Learning, and Human Cognition
Reinforcing the Human Component
Augmenting Clinical Comprehension
Supporting Specific Cognitive Tasks
Mental Models of AI Systems
Conclusion
References
Chapter 6: Machine Learning Systems
Identifying Problems Suited to Machine Learning
The Machine Learning Workflow: Components of a Machine Learning Solution
Evaluating Machine Learning Models: Validation Metrics
Supervised Machine Learning
The Structure of a Supervised Machine Learning Algorithm
Supervised Learning: A Mathematical Formulation
Augmenting Feature Representations: Basis Function Expansion
Bias and Variance
Regularization: Ridge and Lasso Regression
Linear Models for Classification
Discriminative Models: Logistic Regression
Regularized Logistic Regression: Ridge and Lasso Models
A Simple Clinical Example of Logistic Regression
A Multivariate Clinical Example of Logistic Regression
Generative Models: Gaussian Discriminant Analysis
Factored Generative Models: Naive Bayes
Bias and Variance in Generative Models
Recap of Parametric Linear Models for Classification
Non-linear Models
Kernel Methods
Similarity Functions for Kernel Methods
Recap: How to Use Kernels for Classification
Sparse Kernel Machines and Maximum Margin Classifiers
Neural Networks: Stacked Logistic Models
Parameterizing Feedforward Networks and the Forward Propagation Algorithm
Learning the Parameters of a Feedforward Network
Convolutional Networks
Other Network Architectures
Putting It All Together: The Workflow for Training Deep Neural Networks
Ensembling Models
Conclusion
References
Chapter 7: Natural Language Processing
Introduction to NLP and Basic Linguistics Information
Common Biomedical NLP Tasks and Methods
Overview of Biomedical NLP Tasks
Biomedical IE Tasks and Methods
NER Examples and Methods
RE Examples and Methods
CN Examples and Methods
Current Biomedical NLP Tools and Corpora
Biomedical NLP Tools
Biomedical Text Resources
Types of Biomedical Text
Annotated Corpora from Past Challenges
Applications, Challenges and Future Directions
Applications of NLP
Challenges and Future Directions
Conclusion
References
Chapter 8: Explainability in Medical AI
Introduction
Current Trends in AI Explainability Research
Applying Additional Context to Understand Explainability in Medical AI
Three Purposes of AI Explainability
Expanding the Conception of AI Explainability Based on Cognitive Informatics
Human Information Processing
Human-AI Agents
Sociotechnical Systems
Implications of Explainability on Bias and the Regulatory Environment
Explainability and Inherent Biases
Effect of Explainability on Accountability for Decision Making
The Current Regulatory Framework and Explainability
Application of Explainability to Real World Examples of Medical AI
Example: Continuous Blood Glucose Monitoring for Patients with Type 1 Diabetes
Example: Digital Image Analysis Tools Assisting in Histopathological Diagnoses
Example: Wearable Devices Informing Clinical Management
Conclusion
References
Chapter 9: Intelligent Agents and Dialog Systems
Introduction to Dialog Systems
Definitions and Scope
What’s Hard About Getting Machines to Engage in Spontaneous Human Conversation?
Machine Learning and Dialog Systems
History of Dialog Systems in Healthcare
Dialog System Technology
Classic Symbolic Pipeline Architectures
Neural Network Methods and End-to-End Architectures
Approaches to Dialog System Evaluation
Evaluation of Pipeline Architectures
Automated Metrics for End-to-End Architectures
System-Level Evaluation
Example Patient- and Consumer-Facing Dialog Systems
Example Provider-Facing Dialog Systems
Safety Issues in Dialog Systems for Healthcare
State of the Art: What We Currently Can and Can’t Do
Future Directions
Conclusion
References
Part III: Applications
Chapter 10: Integration of AI for Clinical Decision Support
Challenges Faced by Clinicians
Artificial Intelligence-Based CDS
Degree of Automation in AI-CDS
Application of AI-CDS in Clinical Care
Pitfalls of AI-CDS
Regulation of AI-CDS
Conclusions
References
Chapter 11: Predicting Medical Outcomes
Clinical Outcomes: An Enlarged Perspective
AI Approaches for Clinical Outcomes Prediction
Preprocessing: Missing Values, Features Transformation and Latent Variables Extraction
Missing Values
Dimensionality Reduction and Feature Transformation
Deep Learning
Classification
Regression
Survival Analysis
Time Lines and Trajectory Modeling
Markov Models
Performance Assessment
Experimental Design for Learning
Common Mistakes in the Design of Experimental Validation
Experimental Design for Testing: External Validation
Checking Performance Stability, Model Drifts, Diagnostics, and Model Revision
Case Studies and Examples
Type 2 Diabetes
Myelodysplastic Syndromes
The COVID-19 Pandemic
Conclusion
References
Chapter 12: Interpreting Medical Images
Overview
Introduction to Medical Images
Characteristics of Medical Images
Historical Perspectives
Pioneer CAD Systems
Recent Successes in Deep Learning
Clinical Needs and Existing Challenges
Clinical Needs
Medical Applications
Technical Barriers
Opportunities and Emerging Techniques
Acquiring Annotation from Human Experts
Utilizing Annotation by Advanced Models
Extracting Features from Unannotated Images
Conclusion
References
Chapter 13: Public Health Applications
Public Health and AI
Public Health, Essential Public Health Functions, and Public Health Informatics
The Nature of Essential Public Health Functions and the Application of AI
A Vision for AI in Public Health
Applications of AI in Public Health
Examples of AI Applications to Public Health Functions
Assessment
Policy Development
Assurance
Barriers and Risks to AI Applications in Public Health
Future Applications of AI in Public Health
Progress Towards the Vision
Future Applications
References
Chapter 14: AI in Translational Bioinformatics and Precision Medicine
Introduction and Concepts
A Brief History of Translational Bioinformatics
Concepts of AI in Translational Bioinformatics
Primary Data Categories in Translational Bioinformatics
Genomic Data
Clinomic Data
Phenotypic Data
Categorizing AI Applications in Translational Bioinformatics
G2G (Genomic to Genomic)
G2P (Genomic to Phenotypic): Genome-Wide Association Studies (GWAS)
P2P (Phenotypic to Phenotypic): Identify Disease Genomic Subtypes
P2C (Phenotypic to Clinomic)
C2C (Clinomic to Clinomic)
Informatics Challenges in Translational Bioinformatics
Big Data Characteristics
Volume of Data
Veracity of Data
Variability of Data
Velocity of Data
Social-Economic Bias
Domain Knowledge Representation and Interpretability
Model Robustness and Quality Control
Translational Bioinformatics Tools & Infrastructure
Extended Data Management Systems
Data Preprocessing Pipelines
Pipelines to Build the Data Matrix
Enhancing the Data Matrix
Supervised and Unsupervised Learning
Popular Algorithms in Translational Bioinformatics
Classification Algorithms
Clustering Algorithms
Dimension Reduction Algorithms
Association Mining Algorithms
Security, Privacy, and Ethical Considerations (see also Chap. 18)
Team Data Science Infrastructure
Applications of AI in Translational Bioinformatics
Improving Translational Bioinformatics Data Infrastructure
Inferring Pairwise Molecular Regulation
Inferring and Characterizing Cellular Signaling Mechanism that Determines the Cellular Response
Identifying and Characterizing New Cell Types and Subtypes
Drug Repurposing
Supporting Clinical Decisions with Bioinformatics Analysis
Predicting Complex Biochemical Structures
Trends and Outlook
References
Chapter 15: Health Systems Management
Promise of AI in Health Systems
Example: Outpatient Scheduling
Example: Device Monitoring
Governance
Concluding Remarks
References
Chapter 16: Intelligent Systems in Learning and Education
Introduction
Historical Evolution of Medical Education: Philosophical Perspectives and Related Educational Strategies
Acquisition of Clinical Competence
Cognitive Approaches to Learning and Instruction
Approaches to Artificial Intelligence in Education and Training
Artificial Intelligence Systems and the Individual Learner
Computable Representations
Intelligent Tutoring Systems
Dialog Systems and Natural Language Processing
Question Generation
Dynamic Assessment, Feedback, and Guidance
Machine Learning and Neural Networks
Affect and Emotion Aware ITS
Virtual and Augmented Reality
Simulations and Serious Games
Artificial Intelligence Systems and the Education Enterprise
Learning Analytics
Ethics and Regulation
Technology Acceptance and Implementation
Artificial Intelligence Systems in the Future Workplace
The Road Ahead: Opportunities and Challenges for Intelligent Systems in Training, Learning and Practice
References
Part IV: The Future of AI in Medicine: Prospects and Challenges
Chapter 17: Framework for the Evaluation of Clinical AI Systems
The Role of Evaluation: Why It Is Important
Framing Questions for Assessing an Evaluation Plan
Design and Iteration
Cognitive Evaluation Methods
Delivery of Decision Support
Medical Device Data-Interpretation
Event Monitoring and Alerts
Direct Consultation with Clinical User
Naturalistic Studies
Is the System Accepted by Users?
Does the System Have a Positive Impact on User Behavior?
Do Patients Benefit When the System Is Used?
Is Any Positive Outcome Worth the Associated Expense?
Do All Patients Benefit? What Is the Impact on the Population as a Whole?
Additional Considerations
References
Chapter 18: Ethical and Policy Issues
Introduction to the Utility of Applied Ethics
Software Engineering Principles, Standards and Best Practices
Software Engineering of Dependable Systems
Ethics in Good Engineering Practice
Why Context Matters
Trust and Trustworthiness
Explainability and Interpretability
Transparency
The Need for Human Control
Taking the Long View
Fairness and Sources of Bias
Fairness, Bias, Equality, Equity
Data Sets
Algorithmic Design
Implementation and Algorithmovigilance
Organizational and Economic Dimensions
Recommendations for Identifying Bias
Governance and Oversight
AI at Large
AI, Humanity, and Society
AI and the Healthcare Professions
References
Chapter 19: Anticipating the Future of Artificial Intelligence in Medicine and Health Care: A Clinical Data Science Perspective
Introduction
AI in Medicine Technology: An Exponential Rise
Current State
Near Future State
Future State
The AI in Medicine Stakeholders: Increasing Gap to Technology
Current State
Near Future State
Future State
The AI in Medicine Dyads: Synergy and Beyond
The Human-Human Dyad
The Machine-Machine Dyad
The Human-Machine Dyad
Conclusion: Convolution to Consilience
References
Chapter 20: Reflections and Projections
Introduction
Explainability and Complementarity
Restoring Knowledge to AIM
Forward-Thinking Clinical Applications
Workflow, and the Workforce
Evaluation
Concluding Remarks
References
Terms and Definitions
Index