Reconstructing 12-lead ECG from reduced lead sets using an encoder-decoder convolutional neural network

Area of Science:

• Biomedical Engineering
• Cardiovascular Diagnostics
• Signal Processing

Background:

• The 12-lead electrocardiogram (ECG) is the standard for assessing heart electrical activity.
• Current ECG procedures require multiple recording sites, impacting patient comfort and accessibility.
• Reducing the number of leads could streamline monitoring and broaden ECG use.

Purpose of the Study:

• To develop and evaluate a patient-specific framework for reconstructing a 12-lead ECG from a minimal subset of leads.
• To investigate the feasibility of using just three or even a single ECG lead to capture comprehensive cardiac electrical information.
• To simplify ECG monitoring, enhance patient comfort, and increase accessibility in diverse healthcare settings.

Main Methods:

• A patient-specific framework was developed to map multivariate input (subset of leads) to multivariate output (12-lead ECG).
• Preprocessing involved noise elimination and heartbeat segmentation, followed by Short-Time Fourier Transform (STFT).
• An encoder-decoder convolutional neural network (CNN) model was employed for ECG reconstruction.

Main Results:

• The model accurately reconstructed 12-lead ECGs from three leads, achieving average correlations of 97.6% (dataset 1) and 98.9% (PTB Database).
• Reconstruction from a single lead also yielded high accuracy, with average correlations of 97.3% (dataset 1) and 98.4% (PTB Database).
• Correlation coefficients (CC) and root mean square error (RMSE) confirmed the model's effectiveness.

Conclusions:

• The proposed methodology effectively reconstructs 12-lead ECGs using minimal input leads (three or one).
• This approach offers a promising solution for simplified, more accessible, and comfortable cardiac electrical monitoring.
• The findings support the potential for reduced lead ECG systems in various clinical applications.