Research On Automatic Sleep Staging Method Based On Clustering Algorithm

The promotion and development of “World Sleep Day” has also led more and more people to pay attention to their sleep.Medically,sleep is a state of rapid reversibility.Its essential characteristics are the loss of consciousness and the weakening of the response to external stimuli.It plays an important role in the maintenance of human functions.However,diseases caused by sleep have become another major medical challenge.Many sleep-related diseases,through drugs and other treatments,can easily lead to psychophysiological insomnia.What’s worse,it may lead to drug dependence and deterioration.Therefore,effective monitoring and analysis of sleep can help people improve poor sleep quality.Sleep stages are classified according to the different characteristics of electroencephalogram(EEG),electromyogram(EMG),electrocardiogram(ECG),electrooculogram(EOG)and other signals.This paper studies several feature extraction methods and classifier algorithms for sleep EEG,and applies them to sleep staging by analyzing and processing EEG signals.The experimental Data were obtained from EEG Data of 10 normal subjects in the Sleep-EDF database of the open source dataset Physionet Data Bank.The main research contents of this paper are as follows:(1)In view of the low efficiency and redundancy of traditional methods of EEG feature extraction,this paper proposes a method of EEG feature extraction based on Empirical Mode Decomposition(EMD)algorithm and Multi-scale Fuzzy Entropy(MFE).On the one hand,an improved EMD algorithm is proposed based on the complete ensemble empirical mode decomposition with adaptive noise algorithm.An improved Empirical Mode Decomposition model is constructed by defining a "frequency domain correlation coefficient",and feature extraction is achieved by calculating four general feature vectors.On the other hand,this paper uses Multi-scale Fuzzy Entropy as a measure of sleep grading threshold to distinguish different stages of sleep effectively,and constructs a new set of features combined with general feature vectors to realize the characteristics of EEG signals.The experimental results show that the proposed method can effectively improve the recognition rate of EEG features and achieve better automatic sleep stage classification.(2)An automatic sleep stage classification method based on clustering algorithm is proposed to solve the complex problem of sleep EEG data cycle classification.First,through the analysis of correlation coefficient and correlation distance,an improved k-means clustering algorithm on the basis of the original K-means clustering algorithm is proposed;Second,the study is based on the Gaussian Mixture Model(GMM)clustering algorithm combined with thedensity clustering idea to realize the automatic sleep stage classification method.According to the standard of sleep stages,it is divided into five stages: W(Wakefulness),S1(Stage NREM1),S2(Stage NREM2),SS(Slow-wave Sleep)and REM,and the classification efficiency is improved by clustering.It can better achieve automatic sleep stage classification.