Sleep Stage Classification Based on ECG-Derived Respiration and Heart Rate Variability of Single-Lead ECG Signal

In this paper, we present a sleep stage detection algorithm for Wake, Rapid eye movement (REM), and Non-Rapid eye movement (NREM) stages using Heart rate variability (HRV) and ECG-derived respiration (EDR) waveforms extracted from Electrocardiogram (ECG) signal. First, we remove the baseline wander of ECG signal and indicate its R-peaks to form HRV and EDR signals from 30-second ECG intervals. The EDR is extracted using a Neural-PCA based method. In the next step, several features in time and frequency domains are extracted from EDR and HRV signals. Moreover, several features are extracted from mutual information of HRV and EDR, which is derived from cross-spectral and magnitude-squared coherence of both signals. The extracted features were evaluated statistically using the nonparametric Kruskal-Wallis test, and the optimum number of features were selected by the Minimum Redundancy Maximum Relevance (mRMR) algorithm. The sleep stage classification has been done with a multi-class Support Vector Machine (SVM) classifier with Error-Correcting Out-put Codes (ECOC) extension and Radial Basis Function (RBF). The performance of the proposed method was evaluated using the MIT-BIH Polysomnographic database. The obtained accuracy for classification of two classes (Sleep vs. Wake) was up to S1.76%, with Specificity 92.35%, and for three classes, the accuracy was up to 76%, with Specificity 81.39%.