Leadwise clustering multi-branch network for multi-label ECG classification.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Signal Processing

Background:

• The 12-lead electrocardiogram (ECG) is crucial for diagnosing cardiovascular diseases.
• Deep learning models show promise for automated ECG signal classification.
• Current methods often overlook intrinsic lead relationships and cardiac structure in ECG analysis.

Purpose of the Study:

• To develop a deep learning model that better utilizes medical domain knowledge for multi-label ECG classification.
• To introduce an effective lead grouping strategy and a multi-branch network architecture.
• To improve the accuracy of automated ECG classification by integrating features from different leads.

Main Methods:

• Proposed a multi-branch network where each branch uses a multi-scale convolutional structure to extract features from specific lead combinations.
• Introduced a feature weighting fusion module to integrate information from different network branches.
• Evaluated the model on the PTB-XL and CPSC2018 datasets for arrhythmia classification.

Main Results:

• The proposed multi-branch network demonstrated superior performance compared to state-of-the-art methods on multi-label ECG classification tasks.
• The lead grouping strategy and feature fusion module effectively enhanced classification accuracy.
• The model successfully classified multiple arrhythmia types and normal rhythms across different datasets.

Conclusions:

• The novel multi-branch network effectively leverages ECG lead relationships and cardiac structure for improved classification.
• This approach offers a more sophisticated method for automated ECG analysis in clinical settings.
• The findings suggest a promising direction for deep learning applications in cardiovascular diagnostics.