Artificial intelligence-enhanced electrocardiography improves the detection of coronary artery disease

Abstract

An AI-assisted algorithm has been developed to improve the detection of significant coronary artery disease (CAD) in high-risk individuals who have normal electrocardiograms (ECGs). This retrospective study analyzed ECGs from patients aged ≥ 18 years who were undergoing coronary angiography to obtain a clinical diagnosis at Chang Gung Memorial Hospital in Taiwan. Utilizing 12-lead ECG datasets, the algorithm integrated features like time intervals, amplitudes, and slope between peaks, a total of 561 features, with the XGBoost model yielding the best performance. The AI-enhanced ECG algorithm demonstrated high sensitivity (0.82–0.84) when detecting CAD in patients with normal ECGs and gave remarkably high prediction rates among those with abnormal ECGs, both with and without ischemia (92 %-95 % and 80 %-83 %, respectively). Notably, the algorithm's top features, mostly related to slope and amplitude differences, are challenging for clinicians to discern manually. Additionally, the study highlights significant sex differences regarding feature prediction and ranking. Comparatively, the AI-enhanced ECG's detection capability matched that of myocardial perfusion scintigraphy, which is a costly nuclear medicine test, and offers a more accessible alternative for identifying significant CAD, especially among patients with atypical ECG readings.