If this handbook has helped your work or understanding of clinical AI, I would love to hear about it. Anonymous feedback is welcome. Tell me how
Artificial intelligence is entering clinical medicine not as a revolution that will replace physicians, but as a new category of tool requiring the same critical evaluation we apply to any clinical intervention.
Algorithms interpret chest radiographs. Deep learning models analyze pathology slides. Natural language processing extracts information from electronic health records. But published performance does not by itself establish clinical utility in a local workflow. Promotional claims can exceed the independent evidence for the exact product, version, population, and intended use. Even authorized or well-validated systems require careful implementation, explicit monitoring, and honest acknowledgment of limitations.
This handbook addresses the challenge physicians face: How do we critically evaluate these tools, integrate them responsibly into clinical workflow, understand their limitations, and decide which applications genuinely improve patient care?
This is a field guide for evidence-based AI evaluation, not a computer science textbook. If you want to build neural networks from scratch, excellent technical resources exist elsewhere.
This handbook prioritizes peer-reviewed evidence from major medical journals, exact FDA authorization records, and real-world implementations over press releases and vendor whitepapers. Where evidence clearly supports specific tools, I name them. Where products have documented limitations or failures, I describe the evidence and its boundaries. Physicians deserve honest assessments, not diplomatic neutrality between well-supported and unsupported claims.
You don’t need to read sequentially. Jump directly to your specialty chapter. Use the search function. Read the TL;DR summaries for quick orientation. Dive deep when evaluating tools for your practice.
Three approaches:
Every chapter begins with an expandable Chapter Summary (TL;DR) containing clinical context, key evidence, what works, what does not, and the clinical bottom line. Use the summary for orientation, then read the underlying evidence before making an implementation decision.
Stay skeptical. Match the strength of the claim to the study design, and do not treat retrospective validation as proof of clinical benefit. Professional duties remain, while legal responsibility for AI-assisted care depends on the jurisdiction, intended use, institutional controls, and facts of the case. See Liability and Legal Considerations.
Select the pathway that matches your specialty and immediate needs:
“I need practical AI tools for my daily practice”
Start here:
“Radiology, Pathology, Dermatology, Ophthalmology”
Start here:
“General Surgery, Orthopedics, Neurosurgery, OBGYN”
Start here:
“Internal Medicine, Cardiology, Oncology, Neurology”
Start here:
“I work in fast-paced, high-stakes clinical environments”
Start here:
“I care for pediatric and newborn patients”
Start here:
The handbook helps physicians critically evaluate AI tools, integrate them responsibly into clinical workflows, understand their limitations, and determine which applications have credible evidence of improving care or workflow.
Bryan Tegomoh, MD, MPH, is a physician-epidemiologist whose work includes genomic surveillance, public health, clinical evidence appraisal, and evaluation of computational tools under high-stakes conditions.
Three approaches are available: use chapter TL;DRs for a quick scan, read full chapters for implementation planning, or combine the Foundations, relevant specialty, and Implementation chapters for a specialty-focused path.
The handbook prioritizes peer-reviewed research, official regulatory records, professional-society guidance, and documented real-world implementations. Vendor materials and press releases are labeled as such and do not substitute for independent clinical evidence.
This handbook was written by a physician-epidemiologist who spent the pandemic evaluating computational tools under high-stakes conditions. That experience taught a framework for critical evaluation that applies directly to clinical AI. The decisions physicians make now about which tools to adopt will shape medical practice for decades.
Bryan Tegomoh, MD, MPH Berkeley, California January 2025