A Wearable Physiological Detection System to Monitor Blink From Faint Motion Artifacts by Machine Learning Method

• Published in: IEEE sensors journal Vol. 23; no. 21; pp. 26126 - 26135
• Main Authors: Chen, Yeshen, Feng, Yuan, Chen, Zibin, Zhou, Yuanzhi, Yang, Yihan, Ji, Quanyu, Xu, Weiliang, Zhou, Guoyuan, Gao, Tianyu, Li, Xinming
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.11.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Biomedical monitoring; Blink detection; Blood; Classification; classification and regression trees algorithm (CART); Indicators; Machine learning; Monitoring; Monitoring systems; Motion artifacts; photoplethysmography (PPG); Physiology; Regression analysis; Resource utilization; Sensor systems; Sensors; Wearable computers; Wearable technology
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2023.3312975

With the rise of the concept of health care, wearable health monitoring system has also become a topic of great interest. However, it is hard to realize the demands on multimode physiological indicators monitoring and the requirements of wearable, low consumption, and small size at the same time. Thus, it is important to realize multiple indicators detection within a limited detection system. This article proposes a wearable detection microsystem with only a single sensor, which has the potential to realize dual-mode detection of the photoplethysmograph (PPG) signal and blink motion. The PPG signal has been widely utilized in monitoring systems, while blink artifacts always be seen as noise. This work points out the potential application of blink artifacts in wearable monitoring systems and combines it with PPG detection. A classification method based on the first-order differential features was proposed to capture the faint blink artifact signal in the PPG wave. A classification and regression trees (CART) algorithm was further introduced to raise the accuracy of blink artifact classification to 98.5%. Our detection system and algorithm reveal a novel design route to wearable health monitoring system that improves the wearability and resource utilization of the monitoring system by sufficiently mining the signal feature.