Study On Nonlinearity Of Sleep EEG And Detection Of Sleep Apnea Syndrome

Sleep is the most important physiological activity of human beings. With the accelerated pace of modern life and lifestyle changes, the incidence of a variety of sleep disorders is greatly increased. Sleep apnea syndrome(SAS) is a high incidence of sleep disorders, not only is seriously harm to human health and is the independent risk factor of cardiovascular diseases such as hypertension, myocardial infarction, et al, but also bring huge risks to the family and society. Since PSG, as the traditional method for assessment of apnea, is costly, complex and uncomfortable, researchers try to extract the features of SAS by using signal processing method from single or small amount of physiological signal. EEG is the most intuitive parameter to depict the process of sleep which can reflect nerve electrical activity and functional status of the brain. In this study, we use nonlinear methods to study the nonlinear dynamic characteristics of EEG during sleep, and extract the EEG nonlinear dynamics parameters of SAS patients and achieve the automatic detection of SAS patients.Six SAS patients and six healthy persons(Normal) were used as the research subjects. Firstly, the noise removal algorithm of the sleep EEG is studied. Using the wavelet threshold method and independent component analysis(ICA) algorithm to remove the ECG artifact from the EEG. Compare to traditional artifact subtraction algorithm, the ICA significantly improves signal to noise ration and the wavelet de-noising method is fast and effective to suppress the white noise in the brain. Based on the perfect theory and sutable for studying nonlinear sequences of mutual information method, and small amount of data required and high computational efficiency of Cao method, we used the two methods to calculate delay time and embedding dimension respectively to reconstruct phase space of sleep EEG; and then the IAAFT surrogate data method is used to verify the nonlinearity of the data The study confirms that the two groups of EEG signals have chaotic characteristics, which are suitable for the analysis of the nonlinear method.Four nonlinear parameters of the fractal characteristics and the complexity of the sequence are studied:(1) Fractal characteristics of sleep EEG by using correlation dimension. It was found that the correlation dimension of sleep EEG in SAS group and Normal group decreased gradually with the deepening of sleep, which increased to the level of aweak and light sleep stage in REM stage; while the correlation dimension of SAS group was significantly lower than that of Normal group(p<0.01). The correlation dimension is the fractal dimension to describe degree of certainty and regularity of the system. The change of the correlation dimension showed that the activity of brain cells decreased gradually with the deepening of sleep while the significant difference between the two groups hinted that the pathological changes of SAS patients had significant effect on brain nerve activity.(2) A power law dependence of sleep EEG of SAS and Normal groups was studied by using the detrended fluctuation analysis(DFA). DFA is an effective method to study long range correlation in the time series. The results showed that the scaling exponents increased gradually with the deepening of sleep, and in the REM stage, the scaling exponents decreased; and those of SAS were found significantly higher than those of Normal group(p<0.01). It is suggested that with the pathological effects, EEG of SAS patients have a stronger self similarity rule and more smooth oscillation mode than that of healthy people.(3) The sleep EEG is depicted by using the sample entropy(SampEn) and Lempel-Ziv(LZ) complexity respectively from the angle of entropy and symbol dynamics. The two complexity indicators of SAS group and Normal group showed a consistent change trend which conformed closely with the physiological process: with the deepening of sleep, EEG complexity gradually decreased to the lowest, while in the REM stage, complexity increased again; meanwhile, there were significant differences between SAS group and Normal group in both SampEn and LZ complexity(p<0.01) and those in SAS group were significantly lower than those in Normal group. The results suggested that the brain hypoxia or other pathological changes caused by sleep apnea has a significant effect on activity of brain cells and the complexity of brain is decreased.According to the above research, we got the innovative conclusions that the nonlinear characteristics of sleep EEG signal between SAS and Normal were significantly different. It proved the pathological characteristics of SAS have a significant effect on the function and activity of the brain. On this basis, this paper creatively proposed using nonlinear characteristics of sleep EEG for SAS detection, and support vector machine(SVM) algorithm is adopted to classify the SAS patients by using these four characteristic parameters. SVM algorithm is suitable for small sample learning and generalization ability. The results show that the four parameters have a certain degree of identification ability, especially the DFA scaling exponent. The distinguish accuracy of DFA is steadily high and the average is up to 97.2%; as well the accuracy of LZ complexity and SampEn were 82.5% and 83.8% respectively and the sensitivity of LZ complexity was 92.3%; when using combination of these factors to judge, the distinguish accuracy of combination fo DFA and LZ complexity was the highest and the specificity was 96.5%.It is proved that the nonlinear dynamic parameters of the sleep EEG can well characterize SAS and provides a new direction for the study of SAS.