HA-ResNet: Residual Neural Network With Hidden Attention for ECG Arrhythmia Detection Using Two-Dimensional Signal

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Arrhythmia, an abnormal heart rhythm, is a significant clinical challenge.
• Manual electrocardiogram (ECG) interpretation for arrhythmia diagnosis is time-consuming and labor-intensive.
• Deep learning offers potential for automated arrhythmia detection, but feature extraction from noisy ECG signals remains difficult.

Purpose of the Study:

• To propose a novel hidden attention residual network (HA-ResNet) for automated arrhythmia classification.
• To address challenges in extracting Spatio-temporal features from raw ECG signals.
• To improve the efficiency and accuracy of arrhythmia diagnosis.

Main Methods:

• ECG signals converted into 2D images and processed through an embedding layer for shallow feature extraction.
• A hidden attention layer, integrating Squeeze-Excitation (SE) block and Bidirectional Convolutional LSTM (BConvLSTM), captures deep Spatio-temporal features.
• Evaluation on two public datasets using varying segment lengths (2s, 5s, 10s).

Main Results:

• HA-ResNet achieved high F1 scores: 96.0% (2s), 96.7% (5s), and 87.6% (10s).
• The proposed method significantly outperformed existing state-of-the-art approaches.
• Demonstrated effectiveness and generalization capabilities across different data segments.

Conclusions:

• The HA-ResNet model provides an effective solution for automated arrhythmia classification.
• The novel attention mechanism enhances the extraction of critical Spatio-temporal features from ECG data.
• This approach holds promise for improving clinical diagnosis of arrhythmias.