Generative adversarial networks with fully connected layers to denoise PPG signals.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Photoplethysmography (PPG) signals are crucial for non-invasive physiological monitoring.
• Motion artifacts frequently corrupt PPG signals, limiting their clinical utility.
• Existing methods for PPG reconstruction often rely on additional sensor data.

Purpose of the Study:

• To develop and evaluate a novel method for reconstructing motion-corrupted PPG signals.
• To assess the effectiveness of a Fully Connected Generative Adversarial Network (FC-GAN) for PPG signal restoration.
• To determine if PPG reconstruction can be achieved without relying on accelerometer or gyroscope data.

Main Methods:

• A Fully Connected Generative Adversarial Network (FC-GAN) architecture was designed for PPG signal reconstruction.
• Clean PPG signals from the BIDMC Heart Rate dataset were artificially corrupted to simulate motion artifacts.
• The FC-GAN model was trained, validated, and tested using these processed datasets.

Main Results:

• The FC-GAN model demonstrated significant effectiveness in reconstructing distorted PPG signals.
• The reconstructed PPG signals yielded a Mean Absolute Error (MAE) of 1.31 BPM when compared to the ground truth heart rate.
• Performance was evaluated across a heart rate range of 70-115 BPM.

Conclusions:

• The proposed FC-GAN model is robust in reconstructing PPG signals corrupted by varying degrees of motion artifacts.
• The model's accuracy across a typical heart rate range suggests its potential for real-time PPG signal processing.
• This approach offers a promising avenue for PPG reconstruction without the need for auxiliary motion sensors.