Sleep stage detection using a wristwatch-type physiological sensing device

Here, we propose an automated sleep stage classification method using a wristwatch-type physiological sensing device including a reflective photoelectric volume pulse sensor and a three-axis accelerometer to allow simple and inexpensive assessment of sleep quality. One hundred healthy volunteers (60 males and 40 females, aged 20–60 years old) wore the wristwatch-type physiological sensing device during overnight full polysomnography (PSG). Pulse-to-pulse intervals (PPI) and body movement indexes were determined from the records of the sensing device. The features extracted from the time-domain measures of PPI together with wrist movement indexes were utilized to develop an automated sleep stage classification system. The sleep stages detected by the proposed algorithm were compared with those determined by PSG utilizing the standard performance metrics (accuracy, recall, precision and F-measure). The mean rates of agreement with sleep stages on PSG categorized into two stages (wake and sleep), three stages (wake, non-REM, REM), and four stages (wake, light, and deep non-REM, REM) were 87.3, 74.3, and 68.5%, respectively. The accuracy of sleep stage detection indicated that the proposed method was sufficient for assessing sleep quality in healthy subjects without sleep disorders, and the wristwatch-type sensing device developed here can be used as a portable and easy-to-use home sleep monitoring device.