A coordinated adaptive multiscale enhanced spatio-temporal fusion network for multi-lead electrocardiogram arrhythmia...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Multi-lead electrocardiograms (ECGs) are crucial for diagnosing cardiac conditions.
• Automated ECG diagnostic networks using deep learning are increasingly vital.
• Existing models like CNN, RNN, and Transformers have limitations in feature extraction.

Purpose of the Study:

• To propose STFAC-ECGNet, a hybrid deep learning model for enhanced multi-lead ECG analysis.
• To overcome limitations of individual CNN, RNN, and Transformer models in ECG interpretation.
• To improve the accuracy and robustness of automated cardiac diagnosis.

Main Methods:

• Developed STFAC-ECGNet integrating CAMV-RNN, CBMV-CNN, and TSEF blocks.
• CAMV-RNN block enhances spatial-temporal information and global sequence features.
• CBMV-CNN block fuses spatial and channel information via attention mechanisms.
• TSEF block enables multi-scale fusion of spatial and temporal ECG features.

Main Results:

• STFAC-ECGNet demonstrated superior performance on the PTB-XL and CPSC2018 ECG datasets.
• The model outperformed existing state-of-the-art techniques in multiple diagnostic tasks.
• Experimental results highlight the model's robustness and generalization capabilities.

Conclusions:

• STFAC-ECGNet effectively integrates CNN, RNN, and Transformer architectures for advanced ECG analysis.
• The proposed model offers a significant advancement in automated cardiac diagnosis.
• The findings support the clinical utility of STFAC-ECGNet for improved patient care.