In-Silico Modeling of Sleep Stage Classification System Using Covariance Normalization

Abstract

With surge among count of sleeping disorders across the globe and among every strata of society and the non-availability of sleep medicine tools in the backward regions of the third world nations, the need of automated systems arises. This paper introduces an inexpensive, computerized binary sleep stage classification system classifying the rapid eye movement (REM) and non-rapid eye movement (NREM) stages with the usage of electrocardiogram (ECG) and respiratory effort signals (RES). To avail a baseline classification of the sleep stages, support vector machine (SVM) was employed as the backend classifier that uses the heart rate variability (HRV) and respiratory rate variability (RRV) features derived from ECG and RES, respectively. The baseline system developed using linear SVM kernal performed better with performance accuracy of 73.83% , sensitivity of 84.37% and specificity of 30% in totality. The statistical features extracted from the data contain patient-specific variations that are irrelevant for sleep stage classification. As an effort to minimize these variations, covariance normalization (CVN) was performed, and a system is obtained with an absolute overall classification accuracy of 81.30%.