Wrist movement analysis for long-term home sleep monitoring

Abstract In this paper, we present several original methods for classifying sleep stages, including threshold based and k-means clustering based methods. The proposed algorithms use only acceleration data from non-dominant wrist, resulting in a classification into 4-sleep stages (“awake”, “light sleep”, “deep sleep” and “REM (Rapid eye movement)”) for overnight sleep. We validate our methods by referring to the results of “Fitbit” and subjective feedbacks from volunteers on quality of sleep. Our algorithms compute the duration of each sleep stage to evaluate changes in sleep quality between different nights. A method of calculating a sleep score based on the duration of sleep and the duration of each sleep stage is proposed, which facilitates the evaluation of sleep quality by a single score. 5 volunteers were recruited for the tests. Among all the test nights, the proposed algorithm based on k-means clustering shows a superior or equivalent performance compared to the “Fitbit” results. These promising results allow us to consider a new non-intrusive method for users and medical staff to monitor the evolution of sleep quality through long-term follow-up. In addition, to evaluate the performance of our proposed system in terms of sleep stage classification, we use the PSG (Polysomnography) sleep monitoring gold standard to monitor the sleep of one of the volunteers throughout the night in a hospital’s professional sleep laboratory. This experiment shows that the proposed 5km2 (5 iterations of k-means clustering with k=2) method and the threshold method are in good agreement with the PSG results. The accuracy of awake, REM, light sleep and deep sleep detection reaches respectively 0.78, 0.96, 0.75 and 0.97 by the Threshold method. More specifically, both methods we propose show good performance in the detection of awake and deep sleep. This longitudinal monitoring can help to detect abnormal changes in sleep that are usually a sign of a change in health status.