AB-LSTM: Attention-Based Atrial Fibrillation Detection in Short ECG

Atrial Fibrillation (AF), an irregular heart rhythm, poses significant health risks. Early detection is crucial to prevent cardiovascular issues and sudden death. Traditional AF detection depends on engineered features from 12-lead Electro-CardioGrams (ECGs). This work presents three deep learning architectures CNN, CNN + LSTM and CNN + bidirectional LSTM with attention to automatically extract ECG features for AF identification. The China Physiological Signal Challenge 2021 database is utilized for this study. According to experimental findings, our proposed CNN + bidirectional LSTM architecture was the most effective in AF detection with high classification accuracy.