Automatic sleep staging by simultaneous analysis of ECG andrespiratory signals in long epochs
tEEG, EMG, and EOG are very informative signals recorded in polysomnography (PSG) and used for sleepstaging. Their reliable acquisition at home, however, is difficult. In comparison, ECG and thoracic respira-tory (R) signals are easier to record and can be useful in home sleep monitoring systems. The simultaneousutilization of Heart Rate Variability (HRV) and respiratory (R) signals seems a plausible scenario as bothheart rate (HR) and respiration rate (RR) vary during different sleep states. Therefore, we explored thecombined discriminative capacity (accuracy, sensitivity, and specificity) of ECG/R signals in automaticsleep staging. As baseline, we classified the wakefulness, Stage 2, SWS (slow wave sleep) and REM sleepby using a Support Vector Machine (SVM) fed with a set of features extracted from: (a) HRV (34-features),(b) HRV/ECG-Derived Respiration (45-features), and (c) combined HRV/R (45-features) signals. Approach(a) produced reasonable discriminative capacity, while approach (b) significantly improved the classifi-cation; however, the best outcomes were achieved by using approach (c). Then, we enhanced the SVMclassifier with the Recursive Feature Elimination (RFE) method. The classification results were improvedwith 35 out of the 45 HRV/RS-EDR features. In comparison, best results were obtained by combining 27 outof the 45 features derived from HRV/R signals, in which the optimal feature set selected by the SVM-RFEmethod, included a combination of time domain, time-frequency, and fractal features, as well as entropies.Overall, these improvements revealed that it is possible to simplify home monitoring of sleep disordersand achieve high discriminative capacity (accuracy = 89.32%, specificity = 92.88%, and sensitivity = 78.64%)in automatic sleep staging by the exclusive recording of cardiorespiratory signals.