Description
Written to be usable rather than enthusiastic
Most writing on artificial intelligence in medicine is either promotional or dismissive, and neither helps a clinician deciding whether to trust an automated report. These 181 pages take the narrower and more useful question: for each application, what is the actual evidence, and what happens when it is wrong.
Where it currently performs
ECG interpretation, where algorithmic detection of reduced ejection fraction and of atrial fibrillation from sinus rhythm tracings is genuinely ahead of human capability. Echocardiographic measurement and chamber quantification, where automation reduces the inter-observer variability that has always undermined serial comparison. Imaging segmentation. Risk prediction from routinely collected data.
Where it does not
Given equal weight. Performance falls on populations unlike the training data — a point that matters disproportionately outside North America and Europe. Automated reports fail silently rather than obviously. And a model trained on historical decisions reproduces historical bias, including the under-investigation of women in coronary disease.
The responsibility question
The final chapters deal with what a clinician remains accountable for when acting on an algorithmic output, how to document that reasoning, and how to disagree with a model defensibly. This is the section Academy members raise most often and the one with the least published guidance.
Placement
Recommended across all tracks. Members increasingly encounter these tools whether or not they chose them.
PDF, lifetime access, from CardiologyBooks.com.
Reviews
There are no reviews yet.