Multiple electrocardiogram generator with single-lead electrocardiogram.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Monitoring

Background:

• Standard electrocardiogram (ECG) methods use multiple electrodes for brief measurements.
• Wearable devices offer long-term, single-lead ECG monitoring but lack detailed cardiac information.
• Single-lead ECGs do not capture the heart's 3D electrical activity comprehensively.

Purpose of the Study:

• To synthesize multi-lead ECG signals from single-lead ECG data using a generative adversarial network (GAN).
• To overcome the limitations of single-lead ECGs in capturing detailed cardiac electrophysiology.
• To enhance the diagnostic utility of wearable ECG devices.

Main Methods:

• Training generative adversarial network (GAN) models on two independent datasets (PTB-XL and China dataset).
• Evaluating model performance using Fréchet distance (FD) and mean squared error (MSE).
• Conducting two experiments with swapped training and testing datasets to ensure robustness.

Main Results:

• Achieved low Fréchet distance (FD) and mean squared error (MSE) scores, indicating high similarity between synthesized and reference ECGs.
• Experiment 1: FD score of 7.237 and MSE of 0.024.
• Experiment 2: FD score of 8.055 and MSE of 0.011.

Conclusions:

• The proposed GAN-based method successfully synthesizes realistic multi-lead ECGs from single-lead data.
• The low evaluation scores demonstrate the method's potential for improving diagnostic accuracy in wearable ECG devices.
• This technique can significantly enhance the information derived from single-lead ECGs for long-term cardiac monitoring.