A New Study On Automatic Human Sleep EEG Staging Method

Sleep quality plays a vital role in body recovey,memory capacity and resistance to diseases. With constantly increasing pressures of life, variety of sleep-related diseases have become hidden killers to human health. Therefore, more and more domestic and foreign scholars flash into the study of sleep. During sleep, the human body is not a fixed state, but experiencing a few relatively stable stages. Sleep EEG is electrophysiological recording of these states in the brain, so EEG is a very powerful tool for sleep study. Therefore, it has a great application value of sleep EEG scoring for sleep quality assessment and the diagnosis of sleep-related diseases. The traditional sleep staging method mainly depends on the sleep expert's visual analysis with continuous recordings of sleep EEG obtained. However, such approach is quite time-consuming and efficiency is not high.This study is mainly to design a new automatic sleep staging method and evaluate its performance. We applied three kinds of signal complexity operator Co complexity, Kc complexity and Approximate Entropy to study the discriminative ability of EEG signal at different sleep stage. The t test was used to measure statistical significant difference of complexity measure results. All of these can provide theoretical basis for sleep EEG automatic recognition. Based on the non-linear, non-stationary characteristics of EEG signals, we used Hilbert-Huang transform to extract frequency domain features of sleep EEG, combined with C0 complexity, these features can be very good representations of the sleep EEG. A three-layer BP neural network as a general classifier was built to classify these features into an appropriate sleep stage.7 individual Pz-Oz channel sleep EEG from sleep-EDF of MIT-BIT database were employed to test the performance of the proposed method. Experiment results show a high recognition rate for four states, consisting of Awake, Stage 1+REM, Stage 2 and slow wave stage (SWS).The average stage recognition rate of the proposed method at different stage can all achieve above 80%. It is clear that the method is effective in automatic sleep EEG staging. The method will be well used in real-time sleep quality monitoring and sleep-related diseases diagnosis in clinical practice.