Removal of Motion Artifacts From the PPG Signal Using Attentive Generative Adversarial Networks With Dual Discriminator

With the widespread integration of smartwatches and fitness trackers into daily life, photoplethysmography (PPG) signals have emerged as one of the most popular biosignals. However, motion artifacts often affect these signals, diminishing their practical applicability. This study proposes using generative adversarial networks (GANs) to remove these motion artifacts from the PPG signals without additional motion data from accelerometers or gyroscopes. The proposed method was evaluated across several aspects, such as pulse detection, waveform morphology analysis, uniqueness of each generated signal, signal quality enhancement, and heart rate estimation. In addition, the generalization of the proposed method is examined through testing with unseen users, diverse devices, varying sensor locations, and different subject conditions. Although this method has some limitations, including challenges with envelope filtering and dealing with the data from subjects under intense exercise or anesthesia, it provides a foundation for future advancements.