Sleep-wake stages classification and sleep efficiency estimation using single-lead electrocardiogram

• Authors: Mourad Adnane;Zhongwei Jiang;Zhonghong Yan
• Venue: Expert Systems with Applications: An International Journal
• Year: 2012

Detecting sleep-wake stages is of paramount importance in the study of sleep. Conventional methods of sleep-wake stages classification are based on processing physiological signals such as, electroencephalogram (EEG), electrooculogram (EOG) and electromyogram (EMG) that are, mainly, recorded in hospitals using polysomnography (PSG) systems. In this paper, we present an automatic sleep-wake stages classifier method using a single-lead electrocardiogram (ECG). Then, an important sleep quality marker called sleep efficiency is measured using classifier results. The method is based on the extraction, according to three methods, of several features from the heart rate series (RR series). The three methods are, the heart rate variability (HRV) and the detrended fluctuation analysis (DFA) and a method we propose based on the calculation of local energy of detrended profile of RR series; inspired by DFA we call it windowed DFA (WDFA). A subject-specific scheme was adopted, where a part (around 20%) of a subject's data was used to train the classifier and the remaining part (around 80%) is used for the classification, two sets of features were used, i.e., 12 features, initially, and the optimal set of features (10 features in this paper) selected by the support vector machine recursive feature elimination (SVM-RFE) system. The method was tested on the MIT/BIH Polysomnographic Database (MITBPD) using support vector machine (SVM) for training and classification. A mean classification accuracy of 79.31% (12 features, Cohen's kappa value @k=0.41) and 79.99% (10 features, @k=0.43) are reported. Finally, the measure of sleep efficiency was made using the classification results and compared to the actual sleep efficiency values; an average error of 4.52% and 4.64% are reported for the case of 12 features and 10 features, respectively. The method showed good potential for detecting changes in time series and in the sleep-wake stages classification, which prove that an approach with a single ECG lead can be sufficient to estimate an important sleep quality marker which is the sleep efficiency.