1AI Fundamentals for Clinical Researchers

What Is Artificial Intelligence? A Clinician's Primer
Supervised, Unsupervised, and Self-Supervised Learning
Common AI Architectures in Medicine
Foundation Models and Large Language Models in Clinical Research
Matching the Problem to the Approach
What AI Cannot Do: Limitations and Common Misconceptions

2Data: The Foundation of Medical AI

Why Data Quality Matters More Than Model Complexity
How Much Data Do You Need?
Data Annotation and Labeling for Clinical AI
Class Imbalance: When Rare Conditions Dominate Your Research Question
Data Preprocessing and Augmentation
Multi-Site and Multi-Device Data: The Generalization Challenge
De-Identification, Privacy, and Regulatory Compliance for Research Data
Synthetic Data and Data Simulation

3Developing AI Models

The AI Development Lifecycle: From Hypothesis to Deployment
Training, Validation, and Test Sets: Why You Need All Three
Transfer Learning: Standing on the Shoulders of Large Models
Hyperparameter Tuning and Model Selection
Overfitting, Underfitting, and Regularization
Working with an Engineering Team: The Clinician's Role
Tools and Infrastructure: What You Need to Know

4Evaluating Model Performance

5Clinical Validation and Study Design

Technical Validation vs. Clinical Validation: Understanding the Difference
Designing Retrospective Validation Studies
Designing Prospective Clinical Validation Studies
Multi-Site External Validation
Reader Studies: Measuring AI's Impact on Clinical Decision-Making
Defining Clinically Meaningful Endpoints
Reference Standards and Ground Truth in Medical AI

6Regulatory Pathways for Medical AI

7Bias, Fairness, and Responsible AI

Sources of Bias in Medical AI
Demographic Fairness: Performance Across Populations
Generalization Across Devices, Sites, and Geographies
Explainability and Interpretability in Clinical AI
Transparency and Reproducibility
Ethical Frameworks for AI in Healthcare

8Implementation and Clinical Workflow Integration

From Model to Product: The Implementation Gap
Clinical Workflow Analysis: Where Does AI Fit?
Interoperability: DICOM, HL7 FHIR, and Integration Standards
Human-AI Interaction and User Interface Design
Clinician Trust, Adoption, and Change Management
Monitoring AI Performance in Production

9Publishing and Reporting AI Research

Reporting Standards: CONSORT-AI, SPIRIT-AI, TRIPOD+AI, and STARD
Writing the Methods Section for an AI Study
Common Methodological Pitfalls in Medical AI Papers
Sharing Code, Data, and Models
Peer Review of AI Manuscripts: What Reviewers Look For

AAppendices

Glossary of AI and Machine Learning Terms for Clinicians
Recommended Reading and Courses
Regulatory Standards Quick Reference
Checklist: Is Your AI Study Ready for Publication?

Clinical Validation and Study Design

Moving beyond technical performance to demonstrate that an AI model provides real clinical value. This section bridges the gap between model development and clinical evidence.

5.1

Technical Validation vs. Clinical Validation: Understanding the Differencesoon

Why strong performance on a test set is necessary but not sufficient. The hierarchy of evidence for AI: standalone performance, reader studies, comparative studies, and outcome studies.

5.2

Designing Retrospective Validation Studiessoon

How to design rigorous retrospective studies using existing clinical data. Cohort selection, temporal validation, handling missing data, and the limitations of retrospective evidence.

5.3

Designing Prospective Clinical Validation Studiessoon

Study design for prospective AI validation: endpoints, control arms, randomization, blinding, enrichment strategies, and when a prospective study is necessary versus optional.

5.4

Multi-Site External Validationsoon

Why external validation across multiple sites and populations is the gold standard for demonstrating generalizability, and practical approaches to building multi-site collaborations.

5.5

Reader Studies: Measuring AI's Impact on Clinical Decision-Makingsoon

Designing and interpreting reader studies (with-AI vs. without-AI paradigms). Crossover designs, washout periods, reader selection, and measuring the effect on diagnostic accuracy and efficiency.

5.6

Defining Clinically Meaningful Endpointssoon

Moving beyond surrogate metrics to endpoints that matter: diagnostic accuracy against a meaningful reference standard, time to diagnosis, treatment outcomes, and patient-centered measures.

5.7

Reference Standards and Ground Truth in Medical AIsoon

The challenge of defining "truth" when expert clinicians disagree. Consensus panels, adjudication protocols, imperfect reference standards, and their impact on reported model performance.