Evaluation of a new single-channel EEG-based approach for automated identification of sleep stages

Analysis of sleep steps is vital in the detection and treatment of sleep problems like REM sleep disorders and narcolepsy. Automated processing of sleep stages not only makes detection fast but also raises the accuracy of diagnosis. Therefore, the aim of this paper is to present a new automatic method for identifying sleep stages.

In this study, we used the discrete wavelet transform method for feature extraction, a combined system of ant colony technique and neural network for feature selection, and RUSBoost classifier to automatically identify sleep stages. In the recommended approach, the single-channel EEG is decomposed into four levels through discrete wavelet transform, and some statistical features are derived from these levels. Then, utilizing a blend of ant colony optimization as well as neural networks, relevant features are chosen and used as classifier (i.e., RUSBoost) input.

The RUSBoost classifier achieved accuracies of 97.81%, 94.08%, 92.74%, and 92.15% for two-class (sleep and wake), three-class (wake, NREM, and REM), five-class (five sleep stages), and six-class (wake and five sleep stages) states, respectively.

The proposed strategy for automatic detection of sleep steps can raise the speed of detecting sleep stages and even sleep disorders and can be used for big EEG data. However, the proposed approach was only examined on one database and needs to be validated on other databases in future studies.