Machine Learning and AI for Healthcare: Big Data for Improved Health Outcomes

Table of Contents
About the Author
About the Technical Reviewer
Acknowledgments
Introduction
Chapter 1: What Is Artificial Intelligence?
	A Multifaceted Discipline
	Examining Artificial Intelligence
		Reactive Machines
		Limited Memory—Systems That Think and Act Rationally
		Theory of Mind—Systems That Think Like Humans
		Self-Aware AI—Systems That Are Humans
	What Is Machine Learning?
	What Is Data Science?
	Learning from Real-Time, Big Data
	Applications of AI in Healthcare
		Prediction
		Diagnosis
		Personalized Treatment and Behavior Modification
		Drug Discovery
		Follow-Up Care
	Realizing the Potential of AI in Healthcare
		Understanding Gap
		Fragmented Data
		Appropriate Security
		Data Governance
		Bias
	Software
	Conclusion
Chapter 2: Data
	What Is Data?
	Types of Data
	Big Data
		Volume
			Coping with Data Volume
		Variety
			The Internet of Things
			Legacy Data
		Velocity
		Value
		Veracity
		Validity
		Variability
		Visualization
	Small Data
	Metadata
	Healthcare Data—Little and Big Use Cases
		Predicting Waiting Times
		Reducing Readmissions
		Predictive Analytics
		Electronic Health Records
		Value-Based Care/Engagement
		Healthcare IoT—Real-Time Notifications, Alerts, Automation
		Movement Toward Evidence-Based Medicine
		Public Health
	Evolution of Data and Its Analytics
	Turning Data into Information: Using Big Data
		Descriptive Analytics
		Diagnostic Analytics
		Predictive Analytics
			Use Case: Realizing Personalized Care
			Use Case: Patient Monitoring in Real Time
		Prescriptive Analytics
			Use Case: From Digital To Pharmacology
	Reasoning
		Deduction
		Induction
		Abduction
	How Much Data Do I Need for My Project?
	Challenges of Big Data
		Data Growth
		Infrastructure
		Expertise
		Data Sources
		Quality of Data
		Security
	Resistance
	Policies and Governance
	Fragmentation
	Lack of Data Strategy
	Visualization
	Timeliness of Analysis
	Ethics
	Data and Information Governance
		Data Stewardship
		Data Quality
		Data Security
		Data Availability
		Data Content
		Master Data Management (MDM)
		Use Cases
	Deploying a Big Data Project
	Big Data Tools
	Conclusion
Chapter 3: What Is Machine Learning?
	Basics
		Agent
		Autonomy
		Interface
		Performance
		Goals
		Utility
		Knowledge
		Environment
		Training Data
		Target Function
		Hypothesis
		Learner
		Hypothesis
		Validation
		Dataset
		Feature
		Feature Selection
	What Is Machine Learning?
	How Is Machine Learning Different from Traditional Software Engineering?
	Machine Learning Basics
		Supervised Learning
			Unsupervised Learning
			Semi-supervised
			Reinforcement Learning
			Data Mining
			Parametric and Nonparametric Algorithms
	How Machine Learning Algorithms Work
	How to Perform Machine Learning
		Specifying the Problem
			Examples
			Background Information
			Errors in the Data
		Preparing the Data
			Attribute Selection
			Transforming Data
			Aggregation
			Decomposition
			Scaling
			Weightings
			How Much Data Do I Need?
		Choosing the Learning Method
		Applying the Learning Methods
			Should I Code My Machine Learning Algorithm from Scratch?
			Training and Test Data
			Hold-Back Method
			n-fold Cross-Validation
			Monte-Carlo Cross-Validation
			Try Many Algorithms
		Assessing the Method and Results
			Algorithm Accuracy Evaluation
			Bias and Variance
			Bias
			Variance
			Performance Measures
		Optimization
			Improving Results with Better Data
			Should I Choose a Supervised or Unsupervised Algorithm?
			Ensembles
			Problem Distribution
			Implementation Problems
		Reporting the Results
Chapter 4: Machine Learning Algorithms
	Defining Your Machine Learning Project
		Task (T)
		Performance (P)
		Experience (E)
	Common Libraries for Machine Learning
	Supervised Learning Algorithms
		Classification
		Regression
	Decision trees
		Iterative Dichotomizer 3 (ID3)
		C4.5
		CART
	Ensembles
		Bagging
			Random Forest Decision Trees
		Boosting
			Gradient Boosting
			Adaptive Boosting
	Linear Regression
	Logistic Regression
	SVM
	Naive Bayes
	kNN: k-nearest neighbor
	Neural Networks
		Perceptron
		Artificial Neural Networks
	Deep Learning
		Feedforward Neural Network
		Recurrent Neural Network (RNN)—Long Short-Term Memory
		Convolutional Neural Network
		Modular Neural Network
		Radial Basis Neural Network
	Unsupervised Learning
		Clustering
		K-Means
		Association
		Apriori
	Dimensionality Reduction Algorithms
	Dimension Reduction Techniques
		Missing/Null Values
		Low Variance
		High Correlation
		Random Forest Decision Trees
		Backward Feature Elimination
		Forward Feature Construction
		Principal Component Analysis (PCA)
	Natural Language Processing (NLP)
	Getting Started with NLP
	Preprocessing: Lexical Analysis
		Noise Removal
		Lexicon Normalization
		Porter Stemmer
		Object Standardization
	Syntactic Analysis
		Dependency Parsing
		Part of Speech Tagging
			Reducing Ambiguity
			Identifying Features
			Normalization
			Stopword Removal
	Semantic analysis
	Techniques Used Within NLP
		N-grams
		TF IDF Vectors
		Latent Semantic Analysis
		Cosine Similarity
		Naïve Bayesian Classifier
	Genetic Algorithms
	Best Practices and Considerations
		Good Data Management
		Establish a Performance Baseline
		Spend Time Cleaning Your Data
		Training Time
		Choosing an Appropriate Model
		Choosing Appropriate Variables
		Redundancy
		Overfitting
		Productivity
		Understandability
		Accuracy
		Impact of False Negatives
		Linearity
		Parameters
		Ensembles
	Use Case: Type 2 Diabetes
Chapter 5: Evaluating Learning for Intelligence
	Model Development and Workflow
		Why Are There Two Approaches to Evaluating a Model?
		Evaluation Metrics
			Classification
				Accuracy
				Confusion matrix
				Per-class accuracy
				Logarithmic loss
				Area Under the Curve (AUC)
				Precision, recall, specificity, and F-measure
			Regression
				RMSE
				Percentiles of errors
	Skewed Datasets, Anomalies, and Rare Data
	Parameters and Hyperparameters
	Tuning Hyperparameters
	Hyperparameter Tuning Algorithms
		Grid Search
		Random Search
	Multivariate Testing
		Which Metric Should I Use for Evaluation?
		Correlation Does Not Equal Causation
		What Amount of Change Counts as Real Change?
		Types of Tests, Statistical Power, and Effect Size
		Checking the Distribution of Your Metric
		Determining the Appropriate p Value
		How Many Observations Are Required?
		How Long to Run a Multivariate Test?
		Data Variance
		Spotting Distribution Drift
		Keep a Note of Model Changes
Chapter 6: Ethics of Intelligence
	What Is Ethics?
		What Is Data Science Ethics?
		Data Ethics
	Informed Consent
	Freedom of Choice
	Should a Person’s Data Consent Ever Be Overturned?
	Public Understanding
	Who Owns the Data?
	What Can the Data Be Used For?
	Privacy: Who Can See My Data?
	How Will Data Affect the Future?
		Prioritizing Treatments
		Determining New Treatments and Management Pathways
		More real-world evidence
		Enhancements in Pharmacology
	Optimizing Pathways Through Connectivity—Is There a Limit?
	Security
	Ethics of Artificial Intelligence and Machine Learning
		Machine Bias
		Data Bias
		Human Bias
		Intelligence Bias
		Bias Correction
		Is Bias a Bad Thing?
	Prediction Ethics
		Explaining Predictions
		Protecting Against Mistakes
		Validity
		Preventing Algorithms from Becoming Immoral
		Unintended Consequences
	How Does Humanity Stay in Control of a Complex and Intelligent System?
	Intelligence
	Health Intelligence
	Who Is Liable?
	First-Time Problems
	Defining Fairness
	How Do Machines Affect Our Behavior and Interaction
		Humanity
		Behavior and Addictions
		Economy and Employment
	Affecting the future
	Playing God
	Overhype and Scaremongering
	Stakeholder Buy-In and Alignment
	Policy, Law, and Regulation
	Data and Information Governance
	Is There Such a Thing as Too Much Policy?
	Global standards and schemas
	Do We Need to Treat AI with Humanity?
	Employing Data Ethics Within Your Organization
		Ethical Code
		Ethical Framework Considerations
			Collect the Minimal Amount of Data
			Identify and Scrub Sensitive Data
			Compliance with Applicable Laws and Regulations
	A Hippocratic Oath for Data Scientists
	Auditing Your Frameworks
Chapter 7: Future of Healthcare
	Shifting from Volume to Value
	Evidence-Based Medicine
	Personalized Medicine
	Vision of the Future
	Connected Medicine
		Disease and Condition Management
		Virtual Assistants
		Remote Monitoring
	Medication Adherence
	Accessible Diagnostic Tests
	Smart Implantables
	Digital Health and Therapeutics
	Education
	Incentivized Wellness
	AI
		Mining Records
		Conversational AI
		Making Better Doctors
			Optimization
			Diagnosing disease
			Making and rationalizing decisions
			Drug discovery
			3-D printing
			Personalized prosthetics
			Bioprinting and tissue engineering
			Pharmacology and devices
			Education
			Gene therapy
	Virtual and Augmented Reality
		Virtual Reality
		Augmented Reality
		Merged Reality
		Pain Management
		Physical Therapy
		Cognitive Rehabilitation
		Nursing and Delivery of Medicine
		Virtual Appointments and Classrooms
	Blockchain
		Verifying the Supply Chain
		Incentivized Wellness
		Patient Record Access
	Robots
		Robot-Assisted Surgery
		Exoskeletons
		Inpatient Care
		Companions
		Drones
	Smart Places
		Smart Homes
		Smart Hospitals
	Reductionism
	Innovation vs. Deliberation
Chapter 8: Case Studies
	Case Study Selection
	Conclusion
	Case Study: AI for Imaging of Diabetic Foot Concerns and Prioritization of Referral for Improvements in Morbidity and Mortality
		Background
		Cognitive Vision
		Project Aims
		Challenges
		Conclusions
	Case Study: Outcomes of a Digitally Delivered, Low-Carbohydrate, Type 2 Diabetes Self-Management Program: 1-Year Results of a Single-Arm Longitudinal Study
		Background
		Objectives
		Methods
		Results
		Observations
		Conclusions
	Case Study: Delivering A Scalable and Engaging Digital Therapy for Epilepsy
		Background
		Implementing the Evidence Base
		Sensor-Driven Digital Program
		Research
		Project Impact
		Preliminary Analysis
	Case Study: Improving Learning Outcomes For Junior Doctors Through the Novel Use of Augmented and Virtual Reality
		Background
		Aims
		Project Description
		Conclusions
	Case Study: Big Data, Big Impact, Big Ethics: Diagnosing Disease Risk from Patient Data
		Background
		Platform Services
		Medication Adherence, Efficacy and Burden
		Community Forum
		AI prioritization of patient interactions
		Real-World Evidence
		Ethical Implications of Predictive Analytics
		Integration of the IoT
		Conclusions
Technical Glossary
References
Index