PPG2ECGps: An End-to-End Subject-Specific Deep Neural Network Model for Electrocardiogram Reconstruction from...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiovascular Monitoring

Background:

• Electrocardiograms (ECGs) are vital for cardiovascular health assessment but often lack accessibility.
• Photoplethysmography (PPG) signals from wearables show potential for ECG reconstruction.
• Existing methods often require signal alignment, limiting real-world application.

Purpose of the Study:

• To develop an end-to-end, patient-specific deep learning model for direct ECG reconstruction from PPG signals.
• To overcome the limitation of signal alignment required by previous ECG reconstruction techniques.
• To enhance the accessibility of cardiovascular monitoring through wearable technology.

Main Methods:

• Introduction of PPG2ECGps, a novel deep learning neural network based on the W-Net architecture.
• Implementation of a patient-specific approach for direct ECG signal reconstruction from PPG.
• Validation using a dataset of 500 records, comparing reconstructed ECGs against reference ECGs.

Main Results:

• Achieved a mean Pearson's correlation coefficient of 0.977 mV.
• Obtained a mean root mean square error of 0.037 mV.
• Reached a mean normalized dynamic time-warped distance of 0.010 mV.

Conclusions:

• The PPG2ECGps model successfully reconstructs ECG signals from PPG data without requiring signal alignment.
• The model demonstrates high accuracy and reliability in ECG reconstruction.
• This approach holds significant potential for improving patient monitoring and diagnosis in healthcare via accessible wearable devices.