Research Of Sleep Staging Based On EEG Signals

Human beings spend approximately one-third of their lives in sleep. Sleep is animportant physiological process, and it’s the best way to alleviate fatigue and restore energy.Sleep quality is closely related to physical and mental health. Having a good sleep canimprove the efficiency of work and the quality of life. With the increasing pressure of life,sleep problems have affected human health. There are several relatively stable stages insleep, which is called sleep stages. The research of sleep staging is a premise for theassessment of sleep quality. EEG signals can reflect activities of brain cells in differentsleep stages. So the research of sleep EEG signals is the foundation of sleep stagingstudying, sleep quality improving, sleep diseases diagnosing, and it is important ontheoretical significance and clinical application.The main purpose of this paper is to design a novel automatic sleep staging methodbased on EEG signals. Sleep EEG signals are from Sleep-EDF database of MIT-BIH, andthe algorithm is designed in three aspects of signal preprocessing, feature extraction, andfeature classification.1. Wavelet transform was used in preprocessing to remove interference, which isbrought by acquisition device and other physiological signals.2. Entropy, which is the base of feature extraction, can characterize the changes ofrandom signals.Firstly, according to the correspondence between sleep stages and EEG rhythms,wavelet packet decomposition was selected to extract β and δ rhythms from EEG signalsafter contrasting wavelet decomposition and wavelet packet decomposition.Secondly, power spectral entropy with frequency of0.3~35Hz was used to process βand δ rhythms and these characteristic parameters can reflect the change of EEG signals.Finally, multiscale entropy algorithm based on sample entropy with scale of11,12was adopted in processing EEG signals to get auxiliary parameters.3. Back propagation neural network and support vector machine were chosen infeature classification to classify all parameters into four sleep stages. The results canvalidate that the characteristic parameters extracted are valid in sleep staging.The accuracy of sleep staging using a back propagation neural network or a supportvector machine is high and the accuracy using a support vector machine is higher. Power spectral entropy is mathematically simple, and the data of calculation is reduced bymultiscale entropy with scale of11and12, so the processing speed is fast and the runningtime is short. Therefore, the proposed automatic sleep staging method has a better accuracyand real-time performance, and the real-time monitoring of sleep is possible.