Inferring ECG Waveforms from PPG Signals with a Modified U-Net Neural Network.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Wearable devices enable extended cardiac monitoring beyond clinical settings using photoplethysmogram (PPG) and electrocardiogram (ECG) sensors.
• Continuous ECG monitoring via mobile devices is challenging due to user interaction requirements, unlike PPG.
• Diagnosing cardiac anomalies from PPG is limited by the ECG's established role as the gold standard.

Purpose of the Study:

• To propose a method, PPG2ECG, for inferring ECG waveforms from PPG signals.
• To overcome the limitations of current wearable cardiac monitoring by enabling continuous ECG signal acquisition through PPG.
• To facilitate early identification of heart diseases by providing a feasible and accurate ECG monitoring solution.

Main Methods:

• Developed PPG2ECG, a method mapping PPG to ECG signal domains using convolution filters.
• Employed a U-net inception neural network architecture for transforming PPG input to ECG output signals.
• Evaluated the method using personalized and generalized model strategies.

Main Results:

• Achieved mean error values of 0.015 for personalized models and 0.026 for generalized models.
• Demonstrated the feasibility of inferring ECG signals from PPG signals with high accuracy.
• Showcased short distances between inferred and original ECG signals, validating the method's potential.

Conclusions:

• PPG2ECG offers an accurate and feasible approach for continuous ECG monitoring using PPG signals.
• The method overcomes the limitations of existing wearable cardiac monitoring techniques.
• This technology holds significant potential for assisting in the early detection of cardiovascular diseases.