Arrhythmia Classification Method Based on CNN-Attention-BiTransformer

Abstract

This paper proposes a novel arrhythmia classification method combining Convolutional Neural Networks (CNN), Attention mechanisms, and Bidirectional Transformers (BiTransformer). The method aims to improve the accuracy and robustness of arrhythmia detection in electrocardiogram (ECG) signals. Initially, the CNN module extracts local spatial features from raw ECG data, effectively capturing the morphological characteristics of different arrhythmia types. Subsequently, the Attention mechanism is applied to weigh the importance of different segments in the ECG signal, allowing the model to focus on critical features that are most indicative of arrhythmia. Finally, the BiTransformer module processes the feature sequences bidirectionally, capturing both forward and backward dependencies in the ECG signal. This comprehensive approach enables the model to integrate local and global information, enhancing its ability to classify various arrhythmias accurately. Experiments conducted on the MIT-BIH Arrhythmia Database demonstrate that the proposed method achieves state-of-the-art performance, with a significant improvement in classification accuracy compared to traditional methods. The results highlight the effectiveness of combining CNN, Attention, and BiTransformer for arrhythmia classification, offering a promising direction for automated ECG analysis and clinical applications.