Screening of obstructive sleep apnea using higher order statistics of HRV and EDR signals

Abstract

Sleep apnea is a frequent disorder where breathing process is ceased during the sleep and it is found to be a root for cardiovascular problems. In this study, we tend to detect this syndrome solely from nocturnal ECG records. The proposed method is based on higher order spectrum of heart rate variability (HRV) and ECG-derived respiratory (EDR) signals, which extracted from ECG signal. In order to use quadratic phase coupled harmonics information emerging from non-linearities of the HRV and EDR signals, their bispectral features had been employed. Moreover, these features are complemented by time-domain features which can map the signal irregularities. A least square support vector machine (LS-SVM) classifier has been used to detect apneic episodes. The performance of the proposed method is studied using a publicly available database of Physionet. It is shown that the achieved sensitivity, specificity, and accuracy of the presented method were 90.21%, 86.21%, and 88.21%, respectively.