Multiscale Residual Network Based on Channel Spatial Attention Mechanism for Multilabel ECG Classification.

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Electrocardiogram (ECG) analysis is crucial for cardiovascular disease diagnosis.
• Current ECG classification models primarily address single-label problems, limiting clinical utility.
• Real-world ECGs often exhibit multiple concurrent pathologies, necessitating multi-label classification.

Purpose of the Study:

• To develop an advanced deep learning model for accurate multi-label ECG classification.
• To address the limitation of single-label classification in clinical ECG analysis.
• To improve early detection and auxiliary diagnosis of cardiovascular diseases.

Main Methods:

• Proposed a novel multi-scale residual deep neural network (CSA-MResNet) incorporating a channel-spatial attention mechanism.
• Integrated residual networks in a multi-scale approach to capture diverse ECG data features.
• Employed a channel-spatial attention mechanism to focus on critical ECG segments and channels.

Main Results:

• Achieved an average F1 score of 88.2% for multi-label classification of 9 conditions on the CCDD dataset.
• Demonstrated a 1.7% increase in F1 score compared to benchmark models for multi-label ECG classification.
• Attained an average F1 score of 85.8% on the HF-challenge dataset, showing robust performance.

Conclusions:

• CSA-MResNet offers a feasible and effective method for multi-label ECG classification.
• The model aids cardiologists in rapid, early-stage ECG screening.
• Demonstrated generalization performance across different ECG databases.