Predicting atrial fibrillation recurrence after pulmonary vein isolation from ECG data

Abstract

Catheter-based pulmonary vein isolation (PVI) is widely used to treat atrial fibrillation (AF). The procedure isolates the pulmonary vein from the left atrium to suppress irregular heartbeats. If AF recurs, further interventions may be required. This study investigates model-based prediction of AF recurrence using only electrocardiogram (ECG) data. We benchmarked two neural networks for the prediction task. The task-specific convolutional neural network achieved an AUROC of 0.755, slightly exceeding the foundation model (AUROC 0.742). These results confirm earlier findings that ECGs contain predictive information about AF recurrence and they can be extracted by foundation models.