An intelligent Model for EEG Sleep Stages Classification Using Wavelet Transform Based Hybrid Features extraction Model

Abstract

Background: High-quality sleep plays a major role in improving brain development and lifestyle. Electroencephalogram (EEG) signals are the most important signals collected by polysomnography (PSG) used for sleep staging. Manual sleep scoring is a difficult task, however, developing an automatic sleep stage is important to support experts to detect sleep disorders in early stage. Method: In this paper, an automatic single-channel EEG signal sleep stages classification model is proposed. A Discrete Wavelet Transform (DWT) based EEG feature is suggested. Three types of features including entropy, linear, and statistical features are extracted and evaluated to score sleep stages. First, we applied the DWT to each 30-second epoch to decompose the signal into five bands. Then, EEG features are extracted from each band. EEG signals from two datasets named Dreams, and EFD sleep are used to evaluate the proposed model. Results: We interpreted the results using essential statistical criteria. The results showed that the use of combination features improves the sleep classification results. Based on the results, with the Dream dataset, the classification accuracy rate, Kappa coefficient, and F-score were found 0.97,0.92, 0.95 For the second database, we obtained 0.95,0.94, 0.93 for accuracy rate, Kappa coefficient and F-score respectively. Conclusions: We developed a method to score sleep stages that can be used by healthcare providers to identify sleep disorders.
Keywords: DWT,