Single-Lead ECG Arrhythmia Classification Based on Peak-Enhanced Attention Network and Quality-Aware GAN Data...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Single-lead electrocardiogram (ECG) is crucial for wearable atrial fibrillation (AF) screening.
• Noise contamination and data imbalance hinder accurate detection of subtle pathological ECG features like P-waves and f-waves.
• Existing methods struggle with the scarcity of high-quality annotated abnormal ECG data.

Purpose of the Study:

• To develop an end-to-end framework for robust arrhythmia diagnosis using single-lead ECG signals.
• To address data imbalance and noise issues in ECG signal analysis.
• To improve the detection of subtle pathological characteristics in ECGs for wearable devices.

Main Methods:

• Implemented a Quality-Aware Generative Adversarial Network (QA-GAN) with signal quality evaluation and dynamic soft-labeling for high-fidelity minority class sample synthesis.
• Developed a Peak-Enhanced Attention Convolutional Network (PEAC-Net) incorporating a Peak-Enhanced Attention (PE-Att) module with derivative convolutional kernels.
• Integrated 1D multi-scale dilated convolution (DSGC1D) with bidirectional LSTM for capturing both local and global ECG patterns.

Main Results:

• The proposed model achieved an accuracy of 0.902 and a macro-F1 score of 0.880 on the PhysioNet 2017 dataset.
• Demonstrated superior performance compared to state-of-the-art models in arrhythmia diagnosis.
• Exhibited robust data adaptability on the MIT-BIH dataset, confirming its generalizability.

Conclusions:

• The integrated framework effectively mitigates data imbalance and noise in single-lead ECGs.
• The Peak-Enhanced Attention mechanism accurately captures subtle pathological ECG features.
• The model offers a promising solution for reliable arrhythmia diagnosis in wearable devices.