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.

Appendix

To reaffirm the results obtained via the proposed methodology, three additional random forest models were tested. In the first model, a random split of the complete feature dataset was considered. Approximately two-thirds of data were used to construct the model, and the remaining one-third was used for testing. The accuracies reached 90.2, 90.8, 92.0, 93.7 and 97.3 % for the 6- to 2-state sleep stages, respectively. The difference in the accuracy of this model compared to the model reported by Fraiwan et al. [10], who used the same data distribution for training and testing, is 7.8 %.

In the second model, a fifty percent random splitting of data for training and testing sets was considered. The accuracies achieved for the 6- to 2-state sleep stages were 90.1, 91.2, 92.0, 93.7 and 97.2 %, respectively. A direct comparison with Zhu et al. [36], considering their model for training and testing, indicated that the proposed method’s accuracies are 2.6, 2.3, 2.7 and 1.1 % higher than those obtained in the previous study for the 6- to 3-state sleep stages.

In the third model, 35 EEG signals from the 39 available were considered. The remaining 4 EEG signals (recordings from the second night of subjects 00, 01, 02 and 03) were individually used to test the model. The average accuracies reached 88.8, 90.0, 91.7, 93.2 and 97.7 % for the 6- to 2-state sleep stages, respectively. These three models were built with 64 random trees. Among the compared studies, the best results in terms of accuracies for the 3- to 6-state classifications, independent of the chosen approach for assessment, were achieved using the proposed methodology.