Single-channel EEG sleep stage classification based on a streamlined set of statistical features in wavelet domain

Abstract

The main objective of this study was to enhance the performance of sleep stage classification using single-channel electroencephalograms (EEGs), which are highly desirable for many emerging technologies, such as telemedicine and home care. The proposed method consists of decomposing EEGs by a discrete wavelet transform and computing the kurtosis, skewness and variance of its coefficients at selected levels. A random forest predictor is trained to classify each epoch into one of the Rechtschaffen and Kales' stages. By performing a comprehensive set of tests on 106,376 epochs available from the Physionet public database, it is demonstrated that the use of these three statistical moments has enhanced performance when compared to their application in the time domain. Furthermore, the chosen set of features has the advantage of exhibiting a stable classification performance for all scoring systems, i.e., from 2- to 6-state sleep stages. The stability of the feature set is confirmed with ReliefF tests which show a performance reduction when any individual feature is removed, suggesting that this group of feature cannot be further reduced. The accuracies and kappa coefficients yield higher than 90 % and 0.8, respectively, for all of the 2- to 6-state sleep stage classification cases.

Sleep Stages Classification Using Spectral Based Statistical Moments as Features

“…Liang et al [14] employed linear discriminant analysis (LDA) along with the computation of the entropy of EEG signals in multiple scales. Random forest (RF) is fed with Renyi's entropy in Fraiwan et al [13] and statistical moments in wavelet domain in [12], both extracted from a single EEG channel. In [12,13], among many classification algorithms, RF was the best-performing choice.…”

“…In opposition to them, the work of Ronzhina et al [11] rated statistical moments as important parameters in signal analysis at time domain. More recently, Hassan and Bhuiyan [26] and Silveira et al [12] extracted statistical information after decomposing the signals in time-frequency domain, and used them as classification attributes for sleep staging. Nevertheless in [26] the preprocessing stage applied an expensive datadriven Ensemble Empirical Mode Decomposition (EEMD) method for obtaining the frequency bands.…”

“…As done in [6,12,7], the EEG data used in the current study were obtained from the PhysioNet Sleep-EDF (Expanded) database [28,29]. We used all available signals from 20 subjects recorded during two nights, with the exception those signals recorded in the second night from the subject labeled as 13, which were not available in the database.…”

“…According to Ronzhina et al [11], 8 hours of recorded signal demand from 2 to 4 hours of visual analysis. Since the introduction of the R&K standard, a variety of computational methodologies has been proposed [4,12,13,14,11,7] with the objective of supporting automatic identification of sleep stages. Such methods are usually based on: EEG preprocessing, feature extraction and classification.…”

“…Unlike here, in other works [4,11,7], different feature sets are considered depending on the number of stages to be classified. The classification procedure itself is performed by a random forest, which is considered in [12,13] also in the context of brain pattern classification.…”