Transfer Learning with a Pretrained Foundation Model for Atrial Fibrillation and Flutter Detection on Single-Lead ECG

Atrial fibrillation (AF) is a prevalent arrhythmia that increases the risk of stroke and heart failure. Early detection of AF and related conditions, such as atrial flutter (AFL), is challenging due to their transient nature and subtle electrocardiogram (ECG) characteristics. Although deep learning has shown potential for automated ECG analysis, existing models are often constrained by small, task-specific datasets and lack generalization. In this study, we fine-tune a pretrained ECG foundation model using single-lead ECG data for arrhythmia classification. We conduct experiments on multiple publicly available datasets and demonstrate that transfer learning with single-lead input effectively detects AF/AFL events with high accuracy. Furthermore, we specifically address the challenge of class imbalance using weighted loss and demonstrate the model’s robustness across diverse real-world datasets.