Attention-Driven Bi-LSTM Architecture for Identifying Atrial Fibrillation in Short ECG Samples

Abstract

 Atrial fibrillation (AF or A-fib) is characterized by an irregular heart rhythm resulting from erratic and accelerated atrial contractions. An irregular heart rhythm is called an arrhythmia Initially, short episodes of abnormal beating occur, which over time become longer or persistent. This condition significantly impacts human health, and early detection of AF is crucial for preventing various illnesses, including cardiovascular diseases that can lead to sudden death. The 12-lead electrocardiogram (ECG), analysed by cardiologists, is typically used to diagnose AF. Unlike previous research on atrial fibrillation (AF) detection that relied heavily on manual feature engineering, our study explores the use of deep learning algorithms to automatically extract features from ECG signals for AF identification. We propose and evaluate three deep learning frameworks: Simple CNN, CNN incorporating LSTM, and CNN-bidirectional LSTM featuring Attention. Using the "China Physiological Signal Challenge 2021" database comprising 730 records, our experiments demonstrate that the CNN-bidirectional LSTM equipped with Attention model achieves the highest classification accuracy of 98% for AF detection, outperforming other architectures.