MLBF-Net: A Multi-Lead-Branch Fusion Network for Multi-Class Arrhythmia Classification Using 12-Lead ECG

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Automatic arrhythmia detection from 12-lead ECG is crucial for cardiovascular disease prevention and diagnosis.
• Existing methods often concatenate ECG leads, potentially neglecting diverse, lead-specific features.
• This limitation can lead to inadequate information learning for comprehensive 12-lead ECG analysis.

Purpose of the Study:

• To propose a novel Multi-Lead-Branch Fusion Network (MLBF-Net) for enhanced arrhythmia classification.
• To integrate multi-loss optimization for jointly learning both diversity and integrity of multi-lead ECG data.
• To improve information fusion strategies for multi-lead ECG analysis.

Main Methods:

• Developed MLBF-Net architecture with multiple lead-specific branches to capture ECG diversity.
• Implemented cross-lead feature fusion by concatenating branch outputs to learn ECG integrity.
• Employed multi-loss co-optimization across individual branches and the fused network.

Main Results:

• MLBF-Net demonstrated superior performance on the China Physiological Signal Challenge 2018 dataset.
• Achieved an average F1 score of 0.855, indicating top-tier arrhythmia classification.
• The results highlight the effectiveness of the proposed information fusion approach.

Conclusions:

• The MLBF-Net architecture successfully addresses the limitations of previous ECG analysis methods.
• Jointly learning diversity and integrity through multi-loss optimization significantly enhances arrhythmia detection.
• The proposed method offers a promising solution for advanced multi-lead ECG analysis and cardiovascular diagnostics.