Gated CNN-Transformer Network for Automatic Cardiovascular Diagnosis using 12-lead Electrocardiogram

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Engineering

Background:

• 12-lead electrocardiogram (ECG) is crucial for cardiovascular disease (CVD) diagnosis.
• Increasing CVD prevalence necessitates accurate, automated diagnostic methods.
• Deep learning offers potential to reduce subjectivity and enhance ECG diagnostic accuracy.

Purpose of the Study:

• To develop an automated 12-lead ECG diagnostic method for cardiovascular disease.
• To leverage deep learning for improved diagnostic accuracy and reduced human subjectivity.
• To propose a novel network architecture for effective feature fusion.

Main Methods:

• A gated CNN-Transformer network was designed for 12-lead ECG analysis.
• Convolutional Neural Network (CNN) blocks extracted signal embeddings, reducing data complexity.
• A dual-branch Transformer structure captured channel and temporal features in low-dimensional embeddings.
• A gating unit fused features from both branches for final diagnosis.

Main Results:

• The proposed end-to-end deep learning approach achieved an 85.3% overall diagnostic accuracy on the CPSC-2018 12-lead ECG dataset.
• The method demonstrated competitive performance compared to existing deep learning algorithms.
• Feature fusion effectively integrated channel and temporal information for enhanced diagnosis.

Conclusions:

• The gated CNN-Transformer network provides an effective method for automated 12-lead ECG diagnosis of CVD.
• This deep learning approach shows promise for improving the accuracy and efficiency of cardiovascular disease detection.
• The fusion of channel and temporal features is a key factor in the model's performance.