Artificial Intelligence in Medicine: A Clinical Reference to Machine Learning, Diagnostic Validation, and Algorithmic Accountability for Practicing Physicians

A practical guide to working alongside the algorithms already in your hospital. You have almost certainly used one this week. An imaging tool that flagged a nodule. A deterioration score that appeared beside a set of observations. A patient who arrived holding a smartwatch reading. These tools are now part of ordinary clinical work, and most of us were never taught how to think about them. That is the gap this book fills. It asks for no mathematics and contains no code. It assumes you are a competent clinician who wants to use these tools well — knowing which ones have earned your confidence, which need watching, and what to do when the output and the patient in front of you disagree. The Clinical Interrogation Protocol runs through all 25 chapters, applied 75 times to the tools you actually encounter. Four questions, each time: What the model claims What the evidence actually shows Where it breaks What you do at the bedside That last question is the point of the book. Every protocol ends with something you can do in the time a real consultation allows. What you will take away: A clear sense of which applications are genuinely well evidenced — mammography screening support, adenoma detection, closed-loop glucose control — and which are not yet, so your confidence lands where it belongs The ability to read a performance claim properly, including the one question about calibration that tells you whether a number means what it appears to mean Practical familiarity with how tools fail: why silence from a system is not reassurance, why a sparse record reads as a low-risk patient, and why a verification step shrinks unless someone protects it Confidence in conversations you may already be having — with a patient asking about their wearable, a colleague proposing a new system, or a committee reviewing a purchase Coverage runs across radiology, pathology, oncology, cardiology, surgery, critical care, primary care, paediatrics, and public health, with chapters on bias and equity, regulation and liability, and the governance that determines whether a tool helps once it is installed. The book closes with The Clinical Interrogation Library — all 75 protocols gathered and indexed four ways, by chapter, by specialty, by tool class, and by what went wrong. Alongside it sits a master checklist for interrogating any tool the book does not cover, and a 60-item drill with fully explained answers. Each chapter includes board-style review questions with complete rationales. The evidence is current through 2026. The specific tools will change. The questions worth asking will not, and they are the same questions good clinicians have always asked of anything that claims to tell them something about a patient.