HCTG-Net: A Hybrid CNN-Transformer Network with Gated Fusion for Automatic ECG Arrhythmia Diagnosis

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Accurate detection of cardiac arrhythmias from electrocardiogram (ECG) signals is crucial for early cardiovascular disease diagnosis.
• The complex, non-linear nature of ECG waveforms presents significant challenges for traditional analysis methods.
• Existing deep learning models struggle to effectively capture both local waveform morphology and long-range temporal dependencies.

Purpose of the Study:

• To propose HCTG-Net, a Hybrid CNN-Transformer Network with Gated Fusion, for enhanced automatic arrhythmia detection from ECG signals.
• To develop a model that jointly captures localized morphological features and long-range temporal dependencies in ECG data.
• To evaluate the performance of HCTG-Net against existing methods using a standard arrhythmia database.

Main Methods:

• A dual-branch architecture combining a residual Convolutional Neural Network (CNN) for local feature extraction and a Transformer for global temporal context modeling.
• Implementation of a learnable gated fusion mechanism for adaptive integration of features from both CNN and Transformer branches at the per-dimension level.
• Experimental validation using the MIT-BIH Arrhythmia Database.

Main Results:

• HCTG-Net achieved superior performance compared to existing methods, with an overall accuracy of 0.9946 and an F1-score of 0.9711.
• Visualization results demonstrated well-clustered feature distributions, indicating robust feature learning by the model.
• Ablation studies confirmed the synergistic and complementary contributions of the CNN, Transformer, and gated fusion modules.

Conclusions:

• HCTG-Net provides a powerful and adaptive framework for automatic ECG-based arrhythmia diagnosis.
• The model's ability to integrate local and global features offers significant potential for real-time clinical applications and wearable healthcare devices.
• This approach advances the field of automated cardiovascular health monitoring through sophisticated signal processing.