Predicting Sleep Fragmentation using Leg Movements during Sleep

Abstract

Sleep Fragmentation is caused by brief arousals during sleep, it has various health implications and it leads to parasomnia disorders such as daytime sleepiness, nightmares and confusional arousals. Recent advancements in sleep science has lead to finding of correlations between leg movements in sleep (LMS) and arousals. While most (70-80%) arousals have leg movement only limited number of LMS have arousals. In this research, we seek to determine if there are significant features of leg movements which can indicate arousal across very different types of sleepers. This research derives from an in-depth study of sleep data collected at sleep lab using standard polysomnogram and a specialized ankle band which uses capacitor, accelerometer and gyroscope sensors. In this research, we explore the prediction of arousals and sleep fragmentation from LMS by identifying the key features of LMS and combining them to maximize the prediction of fragmented sleep. We applied logistic regression on 17 features such as leg movement duration, Gyroscope activity etc. we get a classification accuracy of 88%, our prediction is better than state of the art with true positive rate as 65% with false positive rate as 4.7%; the idea is to maximize the specificity with reasonable sensitivity.