Sleep stage classification in children using photoplethysmogram pulse rate variability

Human sleep is classified into Rapid Eye Movement (REM) and non-REM sleep. In non-REM sleep, the heart rate and respiratory rate decrease whereas during REM sleep, breathing and heart rate become more irregular. As such, identification of sleep stages by monitoring the autonomic regulation of heart rate is a promising approach. In this study we analysed the standard features of heart rate variability extracted from the pulse oximeter photoplethysmogram (PPG) to identify different sleep stages. The overnight PPG signals were recorded from 146 children with the Phone Oximeter™ in addition to overnight polysomnography. The recordings were divided into 1-min segments and labelled as wake, non-REM and REM based on the event log file of the polysomnography. For each segment, six standard time and frequency domain features of heart rate variability were estimated. Two support vector machine classifiers were separately trained to classify wake from sleep and non-REM from REM sleep. Wake and sleep were classified with an accuracy of 77% and REM and non-REM were classified with an accuracy of 80%.