Data Generation Strategies for Enhanced Atrial Fibrillation Prediction in Pacemaker Patients

Atrial Fibrillation (AF) is a significant concern for patients with pacemakers, as it poses a heightened risk of cardiovascular complications, including strokes and limb ischemia. Accurate prediction of AF in this vulnerable population is crucial for early intervention and prevention of these serious outcomes. However, the scarcity of digitized and labeled electrocardiogram (ECG) data has presented a challenge for Deep Learning based prediction models. In this paper, we undertake a comprehensive exploration of innovative data generation strategies specifically tailored for pacemaker patients’ ECG data. Our study meticulously investigates conditional generation and standard augmentation techniques, systematically augmenting the dataset. Notably, the conditional DeepFake generation emerges as a standout technique, significantly improving data distribution, as evidenced by MMD = 8.10−3 and MMD = 4.10−4. This augmentation approach yields data with acceptable FD, RMSE, and PRD scores when compared to a baseline. These compelling findings offer valuable insights into the enhancement of AF prediction in pacemaker wearers, underscoring the pivotal role of data augmentation in advancing cardiac risk assessment.