PPGANet: Removal of Motion Artifacts from the PPG Signal Using Generative Adversarial Networks

With the emergence of smartwatches and fitness trackers, photoplethysmography (PPG) signal has become more widespread in daily life. However, the signal’s usefulness is still limited by motion artifacts. This study explores the potential of using generative adversarial networks (GANs) to eliminate motion artifacts from the PPG signal without relying on additional sensor data from accelerometers or gyroscopes. Our evaluation methods include PPG pulse detection, heart rate estimation, and waveform morphology. While the proposed method lags slightly behind one state-of-the-art technique that utilized additional sensors in performance, it requires less input signal, making it more beneficial for portable or low-cost devices. As shown in the results, this study can serve as a foundation for future single channel-based algorithms.