Feed

Atrial fibrillation (AF) increases the risk of stroke and heart failure, yet accurate quantification of AF burden in daily life remains difficult. Although smartwatch photoplethysmography (PPG) supports continuous monitoring, complex rhythms and signal noise can impair burden estimation. We developed an AI-enhanced dual-modal framework that combines continuous watch-based PPG (W-PPG) with intermittent single-lead watch-based ECG (W-ECG). A hybrid convolutional neural network-long short-term memory model uses high-fidelity W-ECG segments as dynamic anchors to correct long-term W-PPG classifications. In this prospective validation study, 1,054 patients with AF undergoing catheter ablation (mean age, 62.1 years) were evaluated against patch-based ECG as the reference standard. After ECG-based correction, the system achieved 98.60% sensitivity and 99.27% specificity. The mean absolute percentage error of AF burden decreased by 23.4%, from 1.11% to 0.85%, while the Pearson correlation remained 0.9988. This dual-modal approach offers a scalable and clinically practical solution for long-term AF monitoring, improving burden estimation beyond PPG-only devices without requiring continuous multi-lead ECG. It may support personalized AF management and large-scale cardiovascular screening in real-world settings. (NCT06552468).