Reconstruction of Missing Electrocardiography Signals from Photoplethysmography Data Using Deep Neural Network

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Electrocardiogram (ECG) is crucial for diagnosing heart disease but suffers data loss during long-term monitoring due to sensor detachment.
• ECG monitoring requires user participation, limiting its use for continuous heart activity tracking.
• Photoplethysmography (PPG) offers a low-cost, non-invasive alternative for continuous physiological signal monitoring.

Purpose of the Study:

• To develop and validate a deep neural network model for reconstructing missing ECG signals using PPG data.
• To assess the performance of a WNet-based architecture with an added bidirectional long short-term memory network for ECG signal reconstruction.

Main Methods:

• An end-to-end deep learning model was designed, based on WNet architecture and incorporating a bidirectional long short-term memory network.
• The model was trained and validated using 146 records from the MIMIC III matched subset.
• Performance was evaluated using metrics including Pearson's correlation coefficient, RMSE, PRD, and Fréchet distance.

Main Results:

• The proposed deep learning model achieved a Pearson's correlation coefficient of 0.851.
• The model demonstrated a root mean square error (RMSE) of 0.075 and a percentage root mean square difference (PRD) of 5.452.
• A Fréchet distance (FD) of 0.302 indicated high similarity between reconstructed and original ECG signals.

Conclusions:

• The study confirms the feasibility of reconstructing missing ECG signals using PPG data through deep learning.
• The developed model shows promise for enhancing continuous heart monitoring by addressing data loss issues in ECG.
• This approach could lead to more robust and reliable long-term cardiac monitoring solutions.