Optimizing single-lead ECG axis for AI-based detection of myocardial diseases

Wearable devices enable electrocardiograms (ECGs) outside traditional healthcare settings. While these devices are usually equipped with single-lead ECGs to monitor rhythm abnormalities, recent studies show utility in detecting myocardial diseases when combined with artificial intelligence (AI). However, the optimal single-lead axis for detecting myocardial diseases beyond rhythm abnormality remains unclear. To address this knowledge gap, we trained models to detect 3 myocardial diseases: left ventricular systolic dysfunction (LVSD), hypertrophic cardiomyopathy (HCM), and cardiac amyloidosis (CA) using single-lead ECGs, systematically synthesized at 10-degree intervals, leveraging the limb leads from 12-lead ECG data. For all 3 diseases, the highest discriminations were observed near the aVR axis and its inverse. The optimal angles were 20° for LVSD (AUROC 0.884 [95% CI: 0.877-0.891]) and HCM (0.911 [0.900-0.922]), and 210° for CA (0.893 [0.888-0.899]). These findings may inform the development of wearable ECG devices with utility beyond rhythm abnormality detection.