Sleep Staging Based On Mutiscale Nonlinear Analysis Of EEG Signals

Sleep is the best way to relieve fatigue and restore the spirit. However, with the accelerated pace of life, work pressure increases, more and more people suffer from insomnia and other sleep disorders, serious impacting on people’s health. Research on the characteristics of sleep EEG and automatic sleep staging has important application value and theoretical significance,which will improve the quality of sleep and assist the diagnosis of sleep disorders.First, based on LZ complexity and multiscale analysis, multiscale LZ complexity have been designed. Simulation data analysis showed that the method can overcome the deficiency that mono-scale LZ only measure the complexity of a single frequency characteristics, which can capture multiple frequency components in the series, to better reflect the temporal changes of the signals. Using this method to analyze the sleep EEG signal from Physionet database, the results showed the proposed method captures richer information at the different sleep,verifying to the mono-scale LZ complexity.Secondly, based on the fact that fuzzy entropy entropy with appropriate selection of parameters is more stable than the sample entropy, an improved multiscale analysis method, multiscale fuzzy entropy method was designed. This method combines the advantages of both methods, multiscale analysis has good time-frequency characteristics,and can reduce the amount of computation data; fuzzy entropy algorithm which apply exponential function to fuzzy similarity, can better reflect the complexity of the sleep EEG compared with sample entropy. With multiscale fuzzy entropy method to analyze the actual sleep EEG signal, experimental results verified this method is superior to multiscale sample entropy method.Finally, based on EEG features extracted by the above two nonlinear multiscale analysis method, and EEG power of delta and beta rhythms using wavelet transform,and EOG time-domain features, with the above features as characteristic parameters of sleep staging, support vector machine worked as classifier, with the "one to one" multi-class classification method to implement automatic sleep staging. The results demonstrated thatmultiscale nonlinear analysis method that gets the best stage performance for sleep staging.