Practical intelligent diagnostic algorithm for wearable 12-lead ECG via self-supervised learning on large-scale...

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Cardiovascular disease is a significant global health concern.
• Wearable electrocardiogram (ECG) devices are crucial for continuous monitoring and early detection of transient arrhythmias.
• Intelligent diagnostic systems are increasingly vital for analyzing ECG data.

Purpose of the Study:

• To develop and evaluate a self-supervised learning framework for classifying 60 diagnostic terms from wearable 12-lead ECGs.
• To assess the performance of the model using both offline and online testing.
• To enable real-time intelligent diagnosis and detection of abnormal ECG segments.

Main Methods:

• Collected a large dataset of 658,486 wearable 12-lead ECGs (164,538 annotated).
• Implemented four data augmentation techniques to enhance the dataset.
• Developed a self-supervised learning classification framework for ECG analysis.

Main Results:

• The model achieved an average AUROC of 0.975 and an average F1 score of 0.575 in offline testing.
• During a 2-month online test, the model demonstrated an average sensitivity of 0.736, specificity of 0.954, and F1-score of 0.468.
• The framework successfully recognized 60 different ECG diagnostic terms.

Conclusions:

• The proposed self-supervised learning approach effectively analyzes wearable ECG data for intelligent diagnosis.
• This technology facilitates real-time monitoring and early detection of arrhythmias, improving patient outcomes.
• The system supports clinical cardiologists by identifying abnormal ECG segments for further evaluation.