Automatic arrhythmia detection with multi-lead ECG signals based on heterogeneous graph attention networks

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Automatic arrhythmia detection is crucial for cardiovascular health, typically using multi-lead electrocardiogram (ECG) signals.
• Existing methods often ignore lead correlations and require complex feature extraction, hindering performance and data utilization.

Purpose of the Study:

• To propose a novel multi-lead arrhythmia detection model addressing limitations of existing methods.
• To integrate diverse information and capture intra-lead and inter-lead correlations effectively.

Main Methods:

• A heterogeneous graph attention network was developed to model multi-lead ECG data.
• The model utilizes a dual-level attention strategy to capture interactions within and between leads.
• No feature extraction process is required, simplifying the methodology.

Main Results:

• The heterogeneous graph model successfully integrated diverse information and correlations from multi-lead ECG data.
• The dual-level attention mechanism effectively captured complex inter-lead and intra-lead interactions.
• The proposed model demonstrated significant improvements in multi-lead arrhythmia detection accuracy.

Conclusions:

• The novel heterogeneous graph attention network offers an effective data model for multi-lead ECG analysis.
• The model overcomes the limitations of ignored correlations and complex feature engineering in traditional methods.
• This approach significantly enhances the performance of automatic arrhythmia detection.