Classifying Arousals in Sleep Using Deep Neural Networks

Abstract

An excessive number of arousals fragmenting restful sleep may have a deleterious effect on one's health, to include impaired cognitive function, heart disease and diabetes. Sleep science research has determined the correlation between leg movements during sleep with arousals. In pilot data, derived from a polysomnogram and specialized ankle band, 80-85% of sleep arousals had leg movements, but only 20% of those leg movements were associated with an arousal. In this research, we seek to classify leg movements associated with arousals versus sleep. The accuracy of many machine learning classifiers rely on handcrafted features and explicit domain knowledge. Recent advancements in deep learning methods have demonstrated neural networks as accurate classifiers on raw, multichannel time series data. In this paper, we propose a method that uses neural networks for unsupervised feature learning on leg movement data. When applying a deep neural network on nine channels of leg movement data sampled at 50Hz, we get a classification accuracy of 74%.