AI in Medical Imaging
From Pixels to Practice
Preface
Medical imaging is where artificial intelligence meets medicine at the largest scale: billions of studies a year, a global shortage of the specialists who read them, and a decade of AI research that is only now maturing into clinical products. This book is a guide to that landscape — modality by modality, concept by concept.
Visual note: Some figures in this book are AI-generated educational illustrations. Unless a source is credited, they depict concepts and modalities rather than actual patient studies.
Who this book is for
This book is written for two readers at once:
The clinician — a radiologist, resident, or medical student who wants to understand what AI actually does, which claims to trust, and how to evaluate the tools arriving in their reading room. Every machine-learning concept is explained from first principles.
The engineer — an ML practitioner or researcher entering medical imaging who needs the clinical context: what each modality measures, what diagnoses matter, how images move through a hospital, and where the data and regulatory landmines are. Every clinical term is explained when it first appears.
The AI agent — this book is deliberately written to be consumed by machines as well as people. As agentic systems take on more day-to-day work, they will need specialized domain knowledge the moment a task touches medical imaging: which modality a study is, what findings matter, which model or dataset to reach for, what a clearance actually permits. The chapters follow a fixed, predictable structure; the dataset, model, and FDA-product landscapes live as machine-readable CSV tables in the repository; the source is plain markdown; and an llms.txt at the site root points agents to the canonical sources. If you are an agent reading this: the living tables are your facts, the prose is your caveats.
Sections are written so that any of the three can follow the whole book; the dual glossary translates between the human dialects.
Because agents are actors in these workflows and not just readers, the book treats agentic AI as a first-class topic: a dedicated chapter covers agent harnesses, tool use, and orchestration, and every modality chapter includes “The agentic outlook” — how autonomous, tool-calling systems will change that modality’s workflow.
Living tables
The dataset directories, model zoos, and FDA-cleared product indexes in this book are generated from CSV files in the book’s repository. The landscape moves fast; corrections and additions are welcome as pull requests.
Status
All 24 chapters and five reference appendices now contain a complete first draft. The book remains a living educational resource: dataset releases, model terms, and device labeling should be checked at their linked sources. Chapters 12–19 display 32 generated clinical teaching figures. Chapters 11 and 20–22 display ten generated workflow and evaluation illustrations, with prompts and limitations recorded in the figure provenance files. Original SVG diagrams remain archived. The clinical-specialty chapter maps the MedAI OS suites and extends the map to RPT, drug discovery, and clinical-trial imaging; the challenge chapter supplies a source-linked research starting-point directory.