Multi-channel masked autoencoder and comprehensive evaluations for reconstructing 12-lead ECG from arbitrary...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Cardiology

Background:

• Electrocardiograms (ECG) are crucial for diagnosing cardiovascular diseases (CVD).
• Standard 12-lead ECGs are inconvenient; wearable single-lead ECG devices offer a practical alternative but have an information gap.
• Reconstructing 12-lead ECG from single-lead ECG is essential for leveraging wearable technology.

Purpose of the Study:

• To develop a method for reconstructing 12-lead ECG from arbitrary single-lead ECG signals.
• To introduce a comprehensive evaluation benchmark (ECGGenEval) for assessing the quality of reconstructed ECGs.
• To validate the performance of the proposed Multi-Channel Masked Autoencoder (MCMA) model.

Main Methods:

• Proposed a Multi-Channel Masked Autoencoder (MCMA) for reconstructing 12-lead ECG from single-lead ECG.
• Developed ECGGenEval, a benchmark for signal-level, feature-level, and diagnostic-level evaluations.
• Evaluated MCMA performance using metrics such as Mean Square Error (MSE), Pearson correlation, heart rate variability, and F1-scores.

Main Results:

• MCMA achieved state-of-the-art performance in reconstructing 12-lead ECG.
• Signal-level evaluation showed low MSE (0.0175, 0.0654) and high Pearson correlation (0.7772, 0.7287).
• Feature-level and diagnostic-level evaluations demonstrated high accuracy and reliability, with average F1-scores of 0.8233 and 0.8410.

Conclusions:

• The proposed MCMA effectively reconstructs 12-lead ECG from single-lead ECG, bridging the information gap.
• The ECGGenEval benchmark provides a robust framework for evaluating ECG reconstruction methods.
• This approach enhances the utility of wearable single-lead ECG devices for CVD diagnosis.