Intelligent Analysis and Heartbeat Saliency Map Representation of Postoperative Atrial Fibrillation Recurrence Based on Mobile Single-lead Electrocardiogram

Due to the inconvenience of the current regular in-hospital 12-lead electrocardiogram (ECG) follow-up method for patients undergoing radiofrequency catheter ablation (RFCA), poor medical compliance results in the postoperative population experiencing atrial fibrillation (AF) recurrence after RFCA not being diagnosed in time, thus missing the optimal treatment time. This study investigated the feasibility of predicting AF recurrence using follow-up ECG signals within the blanking period. A total of 170 individuals underwent follow-up and mobile single-lead ECG signal collection after receiving RFCA. We developed a deep modified residual network for intra-patient AF recurrence prediction. In addition, we proposed a novel approach to investigate potential indicators of AF recurrence in ECG signals by generating category-level heartbeat saliency maps. Our model achieved an accuracy of 93.18% for intra-patient AF recurrence prediction, and the saliency maps highlighted the importance of ECG waveform features around the S-peak in predicting AF recurrence. This study provides a new perspective on investigating the relationship between AF recurrence and ECG characteristics within the blanking period and initially validates the feasibility of AF recurrence prediction based on mobile single-lead ECG.