Cardiology: what the evidence shows

The tools in this category analyze tests already used in cardiac care: a 12-lead ECG or a stethoscope recording screened for low ejection fraction, heart sounds checked for murmurs, and coronary CT angiograms analyzed for plaque and for narrowings that limit blood flow. A 2025 meta-analysis of AI models that read the ECG for low left ventricular ejection fraction found pooled sensitivity of 86.9% and specificity of 84.4%. AI-ECG also has outcome evidence beyond screening: in a pragmatic randomized trial of 15,965 hospitalized patients in Taiwan, mortality alerts to physicians from an AI-ECG system not reviewed here were associated with lower 90-day all-cause mortality (3.6% versus 4.3%; hazard ratio 0.83).

A 2024 American Heart Association scientific statement said few AI tools have been shown to improve cardiovascular and stroke care sufficiently to be widely adopted. Open questions for the tools here are whether more diagnoses and fewer invasive tests lead to better outcomes for patients, and whether accuracy holds outside the centers and populations where the algorithms were developed.