A Recurrent Neural Sleep-Stage Classifier Using Energy Features of EEG Signals

This paper presents a recurrent neural classifier to automatically classify sleep stages based on energy features of EEG signals by using only one single EEG channel (Fpz-Cz). The energy features are extracted from characteristic waves of EEG signals which can characterize different sleep stages individually. The recurrent neural classifier takes the energy features extracted on 30s epochs from EEG signals and assigns them to one of the five possible stages: Wakefulness, NREM 1, NREM 2, SWS, and REM. Eight sleep recordings obtained from Caucasian males and females without any medication are utilized to validate the proposed method. Moreover, the feedforward neural network and probabilistic neural network are also presented for comparison the performance of the proposed recurrent neural classifier with the same features. The classification rate of the recurrent neural classifier is better than that of the two neural classifiers. The results demonstrate that the proposed recurrent neural classifier with energy features of the characteristic waves of EEG signals can classify sleep stages more efficiently and accurately using only a single EEG channel.