Automatic Sleep Staging Based On EMD

Sleep, as a complex physiological process, is the important stage for the body recovery, and consolidation. The evaluation of sleep quality is achieved through the sleep staging.Sleep staging by a human expert is a very time consuming task and normally could require hours to classify a whole night recording.Therefor, the automatic staging of sleep EEG has a very important significance.The experimental data used in this paper is come from the Sleep-EDF database of MIT-BIT PhysioBank, and selected one EEG and one EOG for sleep staging. This paper mainly divided sleep into the awake period, NREM 1 period, NREM 2 period, NREM 3/ NREM 4 period (deep sleep), REM (rapid eye movement). This paper proposed a method for sleep EEG staging based upon EMD. EEG signal was decomposed into several intrinsic mode functions (IMF) with the method of empirical mode decomposition(EMD).After that,several effective EEG signal components were separated from intrinsic mode function through the algorithm of Fast-ICA, Effective feature parameters are extracted by using the method of sample entropy and Hilbert-Huang transform, and the classification method based on support vector machine is used to realize the sleep stage.The experimental results show that, sample entropy and Hilbert-Huang transform method can extract effective sleep characteristic value. The sample entropy changes with the change of the degree of sleep in a consistent manner. Specifically, with increase of the depth of sleep, the values of sample entropy decrease gradually, and reach to the minimum in NREM 3 and 4 periods, whereas in REM and NREM 1 periods, the values of sample entropy are approximately the same.By using the method of Hilbert-Huang transform of characteristic value of EEG in different sleep phases has certain difference, is mainly due to the different stages of sleep rhythm. But only use the sample entropy and Hilbert-Huang transform method for the effect of sleep stages, and joined the Hilbert transform method to analyze EEG signals, than using only one or two methods effect is better, the average of 10 samples staging accuracy reached 85.41%.The research results show that the proposed method is ideal for the three aspects of sleep EEG signal de-noising, feature extraction and pattern recognition. Thus,the method of automatic sleep stage proposed by this paper was effective.