MULTI-SENSOR PERIODIC LEG MOVEMENTS DETECTION BY CHARACTERIZING LEG MOVEMENTS DURING SLEEP

Abstract

Periodic leg movement disorder (PLMD) is a prevalent movement disorder during sleep, which leads to poor sleep quality. PLMD is diagnosed by first finding periodic limb movements of sleep (PLMS), which are episodic, involuntary, repetitive movements caused by specific muscle contractions. They usually involve with the lower legs, consisting of extension of the big toe and flexion of the ankle. However, existing diagnose method requires a polysomnography (PSG) study for patient in a sleep lab, which is expensive, time consuming and uncomfortable. In this theses, an in-home wireless wearable solution is introduced for PLMS detection and PLMD diagnose. We present a novel algorithm using machine learning to detect PLMS, based on capacitance, acceleration, and gyroscope sensors data collected from a custom multi-sensor wireless wearable ankle band. Techniques of domain-specific preprocessing, multiple features representation, and different learning models and their ensemble are explored. Moreover, an on-line detection technique is further developed, which can detect PLMS in realtime during sleep. In the evaluation, we collect data from 6 patient subjects (4 adults and 2 children) during real in-sleep-lab studies. Each subject's data is collected from an entire night with complete sleep cycle. With leave-one-subject-out cross validation, our system can achieve an overall PLMS detection accuracy of 92%. The average sensitivity and specificity is 89% and 97% respectively.