Robust Reconstruction of Electrocardiogram Using Photoplethysmography: A Subject-Based Model

Area of Science:

• Biomedical Engineering
• Cardiovascular Physiology
• Machine Learning in Healthcare

Background:

• Electrocardiography (ECG) and photoplethysmography (PPG) are non-invasive cardiovascular monitoring techniques.
• While ECG and PPG cycles correlate, waveform correlations are understudied.
• PPG is more convenient to measure than ECG, yet contains rich cardiovascular information.

Purpose of the Study:

• To propose and validate a subject-specific deep learning model for reconstructing ECG signals from PPG signals.
• To investigate optimal data segmentation strategies for accurate ECG reconstruction.
• To assess the feasibility of deriving detailed cardiovascular insights from PPG alone.

Main Methods:

• A subject-specific deep learning model based on bidirectional long short-term memory (BiLSTM) was developed.
• The model was trained and validated using synchronized PPG and ECG signals.
• Data segmentation into equal 1-minute segments was investigated and optimized.

Main Results:

• The model successfully reconstructed long-duration ECG signals (228s) from shorter PPG recordings (60s).
• Optimal segmentation into 1-minute intervals yielded a high Pearson's correlation coefficient (0.818) between reconstructed and reference ECG.
• Low root mean square error (0.083 mV) and dynamic time warping distance (2.12 mV/s) were achieved.

Conclusions:

• Subject-specific deep learning models can effectively reconstruct ECG from PPG.
• This approach offers a promising, convenient method for comprehensive cardiovascular health assessment.
• The findings pave the way for advanced non-invasive cardiovascular monitoring using PPG signals.