Deep learning model screens for transthyretin amyloid cardiomyopathy using ECG images

Transthyretin amyloid cardiomyopathy (ATTR -CM) is a treatable but underrecognized cause of heart failure, with diagnosis often delayed until advanced disease manifests. This gap is amplified in underserved populations at increased risk for ATTR -CM where access to specialist evaluation and advanced cardiac imaging is limited. Electrocardiograms (ECGs) are ubiquitous and often obtained years before ATTR -CM diagnosis in affected individuals, but conventional interpretation lacks the sensitivity and specificity needed for a practical screening tool. Here, we develop an artificial intelligence model that identifies ATTR -CM directly from widely available images of 12 -lead ECGs. The model achieved an area under the receiver operating characteristic curve (AUROC) of 0.87 (95% confidence interval [CI], 0.82-0.91), with performance maintained across patients with echocardiographic features mimicking ATTR-CM. Performance was consistent and generalizable across 8 multinational validation cohorts with a wide range of prevalences across the US and Europe. Prospective deployment across three screening cohorts spanning older Black and Hispanic adults with heart failure and individuals with prior carpal tunnel syndrome surgery demonstrated clinical applicability with increased risk and plausible screening settings. These findings establish ECG imaging as a scalable entry point for ATTR-CM detection, enabling targeted referral for confirmatory testing and earlier initiation of disease-modifying therapy.