A LSTM-Based Realtime Signal Quality Assessment for Photoplethysmogram and Remote Photoplethysmogram

Haoyuan Gao, Xiaopei Wu, Chenyun Shi, Qing Gao, Jidong Geng; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2021, pp. 3831-3840

Monitoring physiological parameters is very important to access individuals' health status. Recent years, remote photoplethysmogram (rPPG) captured from human face by consumer-level cameras is used to estimate heart rate (HR). However, remote sensing signals are more easily affected by motion artifacts and environmental noise, which make the evaluation results unreliable. In this paper, we propose a long-short term memory network (LSTM) to assess the quality of the PPG(rPPG) signals in real time. This algorithm can also seek out the high quality segments from the ultra-long signals quickly. First, we labeled the PPG data by the combination of three traditional methods. Then, a LSTM network was trained to distinguish between clean signals and noisy signals in the PPG database. Finally, the network from the PPG data was verified in the rPPG data. The results of the experiments show that our method can get the signal quality index in real time, and the high-quality fragments extracted by our method indirectly increase the accuracy of HR evaluation.