Abstract：Sample entropy was used to extract features from Boston Beth Israel sleep EEG data, and sleep stages were investigated. In view of the fact that EEG has weak, non-stationary, and low signal-noise ratio features, which are not easy to extract, the raw EEG signals were preprocessed with wavelet transform to eliminate the noise effectively and the sample entropy was calculated to characterize the sleep stage. Calculating results show that during the periods from wake stage to NREM stage Ⅳ, the values of sample entropy decline regularly, which are in accordance with the notation assessed by experts. The result indicates that the EEG processed by the wavelet transform and sample entropy can accurately reflect the changes of the sleep stages. Compared with the approximate entropy, the method is more accurate and faster so that it can be applied to process non-stationary random signals.
2009, Vol. 30 
