A sleep apnea detection approach based on recurrence plots and convolutional neural networks

Abstract

This paper presents a new sleep apnea detection method based on the combination of recurrence plots (RPs) constructed from heart rate variability (HRV) data and convolutional neural networks (CNNs). RPs is built to present nonlinear dynamics of a complex cardio-respiratory system during sleep apnea reflected by HRV data, which is extracted from the electrocardiogram signals. The information contained in RPs are further extracted by CNNs to classify each RP as normal or apnea. This approach is shown to be good for sleep apnea detection since it can exploit dynamic characteristics of the cardiovascular system of human body during sleep and convert them into feasible features for classification processes. The use of CNNs are meaningful when it requires less domain knowledge for feature extraction and selection. Experimental results show that this newly proposed sleep apnea detection method is better and some other appoachs based on RPs and HRV data in terms of the classification performance and the less complexity of the detection system. This also illustrates the capabilities of this study in real world applications since it is simple, low cost, and easy to implement.