PPG Signal Reconstruction Using Deep Convolutional Generative Adversarial Network.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Wearable Technology

Background:

• Photoplethysmography (PPG) is crucial for non-invasive vital sign monitoring in wearables.
• PPG signals are vulnerable to motion artifacts, compromising data accuracy.
• Existing noise reduction methods have limitations with complex or prolonged signal corruption.

Purpose of the Study:

• To develop an advanced method for reconstructing distorted PPG signals.
• To address the limitations of current PPG signal denoising techniques.
• To improve the reliability of vital sign extraction from wearable sensors.

Main Methods:

• Proposed a deep convolutional generative adversarial network (GAN) for PPG signal reconstruction.
• Utilized temporal information from corrupted signals and preceding data.
• Trained and validated the model using smartwatch-collected data from a home-based health application.

Main Results:

• The proposed GAN method outperformed three state-of-the-art reconstruction techniques.
• Achieved the lowest error rates across various noise durations and signal-to-noise ratios (SNR).
• Demonstrated significant improvement in PPG signal quality.

Conclusions:

• The deep convolutional GAN is effective for reconstructing distorted PPG signals.
• The method enhances the reliability of heart rate and heart rate variability measurements.
• This approach holds promise for more accurate wearable health monitoring.