HADLN: Hybrid Attention-Based Deep Learning Network for Automated Arrhythmia Classification.

Area of Science:

• Artificial Intelligence
• Biomedical Signal Processing
• Cardiology

Background:

• Deep learning models show promise in analyzing electrocardiogram (ECG) data, particularly for detecting atrial fibrillation.
• Traditional deep convolution neural networks face challenges with contextual correlations and gradient dispersion.

Purpose of the Study:

• To propose a hybrid attention-based deep learning network (HADLN) for enhanced arrhythmia classification.
• To address limitations of traditional models by incorporating contextual information and improving gradient flow.

Main Methods:

• The HADLN method integrates Residual Network (ResNet) and Bidirectional Long-Short-Term Memory (Bi-LSTM) architectures.
• An attention mechanism is employed to fuse local and global features, enhancing model interpretability.
• The model was trained and validated using the PhysioNet 2017 challenge dataset for classifying ECG signals into four categories: atrial fibrillation, noise, other, and normal.

Main Results:

• The HADLN method demonstrated significant improvements in classification performance.
• Achieved a precision of 0.866, recall of 0.859, accuracy of 0.867, and F1-score of 0.880 via 10-fold cross-validation.
• The model showed enhanced interpretability through the attention mechanism.

Conclusions:

• The proposed HADLN method offers a robust approach for arrhythmia classification using ECG data.
• The fusion of ResNet, Bi-LSTM, and attention mechanisms leads to superior performance and interpretability.
• This AI-driven method holds potential for advancing automated ECG analysis and atrial fibrillation detection.