The Clinical AI Field Guide

Author

Rafe McBeth

Published

January 19, 2026

Welcome

The Clinical AI Field Guide

This book trains you to build clinical AI systems and ship them responsibly.

Models are the easy part. Deployment, governance, monitoring, and the field guide that tells a real clinic how to use your tool safely—that’s the hard part. We cover both.

Part One (Chapters 1–11) teaches you to build: programming foundations, statistics, classical ML, deep learning, medical imaging, NLP, and large language models.

Part Two (Chapters 12–22) teaches you to ship: real-world applications, deployment pipelines, interpretability, privacy, fairness, regulation, and writing the documentation that makes clinical AI safe to use.

The goal is to produce people who can build a model, validate it rigorously, deploy it into a clinical workflow, and write the field guide someone else could use. That’s the skill set that’s missing—and the skill set this course develops.

It accompanies the MPHY 6120 course at the University of Pennsylvania, but stands alone for anyone ready to move from consuming AI tools to building and shipping them.

Who This Book Is For

Clinicians, researchers, and students who want to build and ship clinical AI—not just evaluate vendor products
Data scientists entering healthcare who need clinical context for responsible deployment
Biomedical engineers and medical physicists bridging technical and clinical worlds
Anyone who will be asked to use AI tools they didn’t design, and wants to understand them deeply enough to govern them responsibly

Reader Tracks

Not everyone needs every chapter. Choose your path:

Builder Track — Building and shipping clinical AI systems:

Read sequentially—each chapter builds on the previous
Chapters 2–5 are prerequisites; don’t skip them (especially Ch 2 on problem framing and Ch 5 on data quality)
Code examples are designed to be copied and adapted
The second half (deployment, governance, field guides) is as important as the first

Foundations Track — Not ready to code yet? Start here:

Ch 1–2 (History, Framing AI Problems): Understand the landscape
Ch 5 (EDA): Learn why data quality determines everything
Ch 11 (LLMs), Ch 15 (Deployment), Ch 17–21 (Responsible AI): Understand what it takes to ship safely
Come back for the technical chapters when you’re ready to build

Quick Reference — Evaluating or governing an AI system today:

Ch 17 (Interpretability): What questions to ask about model behavior
Ch 19 (Fairness): Bias red flags and equity considerations
Ch 20 (Regulation): Compliance requirements and FDA pathways
Ch 21 (Writing the Field Guide): Documentation standards for safe deployment

How to Use This Book

Each chapter includes:

Clinical Context: Real-world motivation for why this topic matters
Core Concepts: Theory explained accessibly with medical examples
Code Examples: Executable Python/PyTorch snippets you can adapt
Key Terms: Indexed terminology for quick reference

The code examples use standard libraries (PyTorch, scikit-learn, MONAI, HuggingFace) and are designed to run on modest hardware.

Acknowledgments

This book draws on materials developed for the Penn AI for Medicine course. Thanks to the students and teaching assistants who provided feedback on early drafts.

License

This work is licensed under CC BY-NC-SA 4.0. You are free to share and adapt the material for non-commercial purposes with attribution.

# Welcome {.unnumbered} This book trains you to **build** clinical AI systems and **ship** them responsibly. Models are the easy part. Deployment, governance, monitoring, and the field guide that tells a real clinic how to use your tool safely—that's the hard part. We cover both. **Part One (Chapters 1–11)** teaches you to build: programming foundations, statistics, classical ML, deep learning, medical imaging, NLP, and large language models. **Part Two (Chapters 12–22)** teaches you to ship: real-world applications, deployment pipelines, interpretability, privacy, fairness, regulation, and writing the documentation that makes clinical AI safe to use. The goal is to produce people who can build a model, validate it rigorously, deploy it into a clinical workflow, and write the field guide someone else could use. That's the skill set that's missing—and the skill set this course develops. It accompanies the MPHY 6120 course at the University of Pennsylvania, but stands alone for anyone ready to move from consuming AI tools to building and shipping them. ## Who This Book Is For - Clinicians, researchers, and students who want to **build and ship** clinical AI—not just evaluate vendor products - Data scientists entering healthcare who need clinical context for responsible deployment - Biomedical engineers and medical physicists bridging technical and clinical worlds - Anyone who will be asked to use AI tools they didn't design, and wants to understand them deeply enough to govern them responsibly ## Reader Tracks Not everyone needs every chapter. Choose your path: **Builder Track** — Building and shipping clinical AI systems: - Read sequentially—each chapter builds on the previous - Chapters 2–5 are prerequisites; don't skip them (especially Ch 2 on problem framing and Ch 5 on data quality) - Code examples are designed to be copied and adapted - The second half (deployment, governance, field guides) is as important as the first **Foundations Track** — Not ready to code yet? Start here: - Ch 1–2 (History, Framing AI Problems): Understand the landscape - Ch 5 (EDA): Learn why data quality determines everything - Ch 11 (LLMs), Ch 15 (Deployment), Ch 17–21 (Responsible AI): Understand what it takes to ship safely - Come back for the technical chapters when you're ready to build **Quick Reference** — Evaluating or governing an AI system today: - Ch 17 (Interpretability): What questions to ask about model behavior - Ch 19 (Fairness): Bias red flags and equity considerations - Ch 20 (Regulation): Compliance requirements and FDA pathways - Ch 21 (Writing the Field Guide): Documentation standards for safe deployment ## How to Use This Book Each chapter includes: - **Clinical Context**: Real-world motivation for why this topic matters - **Core Concepts**: Theory explained accessibly with medical examples - **Code Examples**: Executable Python/PyTorch snippets you can adapt - **Key Terms**: Indexed terminology for quick reference The code examples use standard libraries (PyTorch, scikit-learn, MONAI, HuggingFace) and are designed to run on modest hardware. ::: {.content-visible when-format="pdf"} ## A Note on This Edition This print edition focuses on concepts and clinical context—readable, portable, light on code. Full executable code examples with syntax highlighting are available in the online version at [book.clinicalai.guide](https://book.clinicalai.guide). ::: ## Acknowledgments This book draws on materials developed for the Penn AI for Medicine course. Thanks to the students and teaching assistants who provided feedback on early drafts. ## License This work is licensed under [CC BY-NC-SA 4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/). You are free to share and adapt the material for non-commercial purposes with attribution.