Automatic Classification Method of Arrhythmias Based on 12-Lead Electrocardiogram.

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Cardiovascular diseases, particularly arrhythmias, are a leading global cause of mortality.
• Electrocardiograms (ECGs) are crucial for diagnosing arrhythmias, but effectively utilizing 12-lead data remains challenging.
• Existing automated methods often overlook frequency domain features and struggle with information fusion across leads.

Purpose of the Study:

• To develop a highly accurate and generalizable automated arrhythmia detection algorithm using 12-lead ECGs.
• To address limitations in current methods by integrating multimodal feature extraction and fusion.
• To improve the classification of various arrhythmia types.

Main Methods:

• A dual-channel deep neural network was employed to extract features from both 1D ECG sequences and 2D time-frequency representations.
• An attention mechanism was integrated to effectively fuse critical information from all 12 leads.
• The model was trained and validated on a mixed dataset encompassing nine arrhythmia types.

Main Results:

• The developed model achieved an average F1 score of 0.85 and an average accuracy of 0.97.
• The multimodal feature fusion approach successfully integrated diverse ECG signal characteristics.
• Experimental results demonstrated stable and reliable performance across different arrhythmia classifications.

Conclusions:

• The proposed model offers a promising advancement in automated arrhythmia detection.
• Effective fusion of multimodal ECG features, including time and frequency domains, enhances diagnostic accuracy.
• The algorithm shows significant potential for practical clinical application in cardiovascular disease management.