A multi-modal approach to sleep-wake classification in infants using minimally invasive sensors

In this study, we evaluate the potential and efficiency of a low-cost and minimally invasive means of identifying sleep/awake patterns in infants using a combination of pulse-oximetry, electrocardiogram and actigraphy data. Full overnight polysomnogram data from 402 infants from four distinct screening categories was extracted from the National Collaborative Home Infant Monitoring Evaluation (CHIME) database along with hand-scored sleep state annotations and was used to train and validate a classifier model based on linear discriminants. Results for each screening condition are provided along with the overall results across the entire dataset. The overall classifier achieved an accuracy of 74.1%, a sensitivity of 60.9% and a selectivity of 82.0%.