腕時計型ウェアラブル機器による睡眠時生体計測と機械学習による睡眠の多面的評価

抄録

Sleep is an important rest for life support, and it is known that chronic sleep deprivation adversely affects physical and mental health. Therefore, it is considered that it contributes to the maintenance and promotion of physical and mental health by evaluating daily sleep quality and promoting the improvement of sleep habits. In recent years, polysomnography performed at medical institutions is difficult to perform daily at home. In this study, we developed a wristwatch-type wearable device for the purpose of evaluating daily sleep quality and attempted to estimate sleep stage. As a result, accuracy of HMM was 68.1% when the HMM was trained with time-dependent transition probability and outputs of RNN was trained with wrist acceleration and photoplethysmography. In addition, there was a correlation between sleep variables of PSG such as sleep efficiency and sleep latency and the OSA-MA overall score which assesses subjective sleep quality. In the binary evaluation of sleep quality using the standardized average score as a threshold, accuracy of the wearable device and PSG were similar and exceeded 90%. This suggests that it is possible to evaluate sleep quality by measuring daily sleep with the wearable device.