AI-enhanced ECG with segment-specific concept based on coronary artery circulation and ischemic changes: advancing coronary artery disease screening

Abstract Background/Introduction Coronary artery disease (CAD) remains the predominant cause of mortality worldwide, often eluding detection until its advanced stages. Traditional diagnostic methods are both costly and invasive. Purpose This study evaluates the effectiveness of an artificial intelligence (AI)-enhanced electrocardiogram (ECG) that focuses on segment-specific ischemic changes and repolarization abnormalities in improving diagnostic accuracy for CAD. Methods A cohort of 4,218 CAD patients provided 2,023 and 3,206 ECGs from July 2020 to June 2022, which were used to train, validate, and test an AI model in two phases. The data was divided into training, internal validation, and testing sets with an 8:2 and 8:1:1 ratio respectively. Our approach integrates AI to identify and analyze segment-specific repolarization abnormalities and coronary circulation patterns. Results Employing full-length ECG data and targeted ST segment extraction, the AI model distinguished between normal and ≥70% coronary artery stenosis with an accuracy of 0.788 (95% CI: 0.771 – 0.805). It demonstrated precision of 0.767 and 0.811, recall of 0.821 and 0.755, F1-scores of 0.793 and 0.782, and an AUROC of 0.860 for normal and stenosis cases respectively, derived from 1,031 normal and 1,050 stenosis cases. Conclusion AI-enhanced ECG analysis targeting segment-specific ischemic changes holds significant promise for the non-invasive screening of CAD. This methodology could potentially refine the accuracy of CAD screening, decreasing dependency on more invasive diagnostic techniques.