AI detects cardiac amyloidosis from heart tracing images
The model achieved a score of 0.87 for distinguishing cases from noncases in an internal test set of 20,222.
Limit: The study evaluated detection performance but did not assess whether screening improved patient health outcomes
Edited by Boris Djordjevic · AgingResearch.News
In an observational study spanning eight cohorts with 4,276 cases of transthyretin amyloid cardiomyopathy, a form of heart disease, researchers tested an artificial intelligence model using electrocardiogram images—recordings of the heart’s electrical activity. They developed the model in a US hospital system.
The discrimination score measured how well it distinguished cases from noncases; 1 represents perfect separation. It scored 0.89 on 3,902 images from a UK amyloidosis center. Prospective screening cohorts totaling 1,017 participants yielded scores of 0.76 in older Black adults with heart failure, and 0.79 and 0.91 in separate cohorts including people who had carpal tunnel surgery. In broader clinical cohorts, the model performed similarly to a heart ultrasound-based risk score while requiring fewer confirmatory tests per correctly detected case.
Why it matters
For older adults with heart failure, the work addresses whether routine tests could help identify who needs further testing for an underdiagnosed, increasingly treatable heart disease.
Caveats
This observational validation assessed detection performance, not whether screening improved health outcomes. The report is a conference abstract; its prospective screening populations had heart failure or prior carpal tunnel syndrome.
The paper
Yale New Haven Hospital
ESC Congress 2026 Abstract, Oct 26, 2026, Open access
Presented at ESC Congress 2026
