Classification feasibility test on multi-lead electrocardiography signals generated from single-lead...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Wearable devices enable real-time Electrocardiogram (ECG) monitoring but often provide limited data.
• Generating detailed 12-lead ECG signals from limited wearable data is crucial for enhanced cardiac diagnostics.
• Traditional methods for ECG analysis may lack the granularity needed for precise condition identification.

Purpose of the Study:

• To investigate the feasibility of generating detailed 12-lead ECG signals from single-lead (Lead I) data.
• To evaluate the performance of generated 12-lead ECG signals in classifying various cardiac arrhythmias and conditions.
• To compare the diagnostic accuracy of generated 12-lead ECGs against real 12-lead ECGs.

Main Methods:

• Utilized a U-net-based Generative Adversarial Network (GAN) trained on ECG data from Asan Medical Center to synthesize 12-lead ECGs from Lead I.
• Employed unseen PTB-XL PhysioNet data to generate reference real 12-lead ECG signals.
• Compared classification performance using a ResNet model on both generated and real 12-lead ECG datasets for conditions including normal sinus rhythm, atrial fibrillation (A-fib), left bundle branch block (LBBB), right bundle branch block (RBBB), left ventricular hypertrophy (LVH), and right ventricular hypertrophy (RVH).

Main Results:

• The generated 12-lead ECG signals achieved superior classification performance.
• Mean precision, recall, and F1-score for generated signals were 0.82, 0.80, and 0.81, respectively.
• Real 12-lead ECG signals yielded mean precision, recall, and F1-score of 0.70, 0.72, and 0.70, respectively.

Conclusions:

• Generated 12-lead ECG signals demonstrate higher accuracy in classifying cardiac conditions than real 12-lead ECGs.
• This GAN-based approach offers a promising method for enhancing diagnostic capabilities using data from simpler wearable devices.
• The findings suggest a potential for improved remote cardiac monitoring and diagnosis through advanced signal generation techniques.