Coronary artery disease remains a leading global killer, yet doctors have no cheap screen for dangerous blockages before imaging. A model described August 18 in npj Digital Medicine aims to change that with nothing but the standard electrocardiogram. The multicenter study fine-tuned an existing ECG foundation model against CT angiography as the reference standard.
The AI predicts vessel-specific stenosis, defined as 70% or more narrowing in the major arteries and 50% or more in the left main. It held up in internal and external cohorts, in clinically normal ECGs and across demographic subgroups.
Model outputs rose in step with stenosis severity and were grouped into low, intermediate and high risk strata. Combined with guideline-based pretest probability, the model improved risk reclassification, strengthened rule-out performance and shrank the gray zone where doctors order extra tests. Longitudinal follow-up showed model-defined risk groups separated clearly on major adverse cardiovascular events.
Why it matters: ECGs are cheap, ubiquitous and already routine, while CT angiography is costly and exposes patients to radiation. A trustworthy ECG screen could route only the right patients to advanced imaging.
