Enhanced Atrial Fibrillation (AF) Detection via Data Augmentation with Diffusion Model

Atrial Fibrillation (AF) is the most common type of cardiac arrhythmia, posing significant health risks such as strokes and cardiac failure. Deep learning methods have been employed to automatically diagnose AF from electrocardiogram (ECG) signals. In this study, transforming the ECG signal into two-dimensional Poincaré plots, we explore the application of diffusion model in the augmentation of AF samples to improve the performance of AF detection via deep learning. In this regard, we compare the performance of diffusion model and generative adversarial network (GAN). Our results suggest that diffusion model can potentially generate more satisfactory samples in terms of both the FID score and the downstream classification task.