Sleep stage classification using hydraulic bed sensor

Abstract

Sleep monitoring can help physicians diagnose and treat sleep disorders. Polysomnography(PSG) system is the most accurate and comprehensive method widely used in sleep labs to monitor sleep. However, it is expensive and not comfortable, patients have to wear numerous devices on their body surface. So a non-invasive hydraulic bed sensor has been developed to monitor sleep at home. In this thesis, the sleep stage classification problem using hydraulic bed sensor was proposed. The sleep process divided into three classes, awake, rapid eye movement (REM) and non-rapid eye movement (NREM). The ground truth sleep stage came from regularly scheduled PSG studies conducted by a sleep-credentialed physician at the Sleep Center at the Boone Hospital Center (BHC) in Columbia, Missouri. And we were allowed to install our hydraulic bed sensors to their study protocol for consenting patients. The heart rate variability (HRV) features, respiratory rate (RV) features, and linear frequency cepstral coefficient(LFCC) were extracted from the bed sensors' signals. In this study, two scenarios were applied, put all subjects together and leave one subject out. In each scenario, two types of classification structures were implemented, a single classifier and a multi-layered hierarchical method. The results show both potential benefits and limitations for using the hydraulic bed sensors to classify sleep stages.