Research On Multi-scale Entropy Analysis Methods Of EEG

With the continuous development and improvement of brain science,EEG signal has become one of the most commonly used methods for clinical and researchers to diagnose brain diseases.Because of the nonstationarity and non-linearity of EEG,the analysis methods of EEG were still in research stage,and the method of artificial judgment was still used in clinical diagnosis.At present,features of most nonlinear analysis methods for EEG were extracted at a single scale,while the multi-scale characteristics of EEG are neglected.Therefore,the study of multi-scale entropy analysis of EEG has a certain theoretical value.In this paper,two new indexes of variance and standard deviation were introduced,and a multi-scale entropy method based on non-uniform time window was proposed by analyzing the characteristics of the traditional multi-scale entropy methods.The indexes and time windows used during multi-scale process were studied,and the application of three indexes in multi-scale entropy methods was analyzed and compared.According to the characteristics of uniform time windows in traditional multi-scale entropymethod,a new time window division method was proposed and the effectiveness was verified in the epilepsy data,alcohol data and schizophrenia data.The main work are as follows:(1)Two indexes of variance and standard deviation were introduced to multi-scale time series.This paper analyzes the problem of information loss in the traditional multi-scale entropy method,and introduces the variance and standard deviation to analyze the EEG signal.Through the validation in the three data sets,it is found that the use of new indicators can yield better results than the average.(2)Performance of sample entropy and fuzzy entropy in multi-scale entropy methods was research.Sample entropy and fuzzy entropy were selected during extracting features,and the results show that two features can be used for the analysis of EEG signals better,but the performance of fuzzy entropy is better than that of sample entropy and the time complexity of sample entropy is lower than the fuzzy entropy.(3)A multi-scale entropy method with nonuniform time windows was proposed.Focused on the problem of low accuracy in coarse-grained method and high time complexity in fine-grained method,a multi-scale entropy method using nonuniform time windows was proposed according to the idea of nonuniform segmentation based on spatial trend.The feasibility and effectiveness of the method with average,variance and standard deviation three indexes and sample entropy,fuzzy entropy two features were verified in threedata sets.And results show that the nonuniform time window method can obtain better accuracy and lower time complexity.In this paper,the variance and standard deviation were introduced into the traditional multi-scale entropy method with uniform time windows,and the multi-scale entropy method with nonuniform time windows was proposed.Finally,these methods were validated in three data sets.The experimental results show that more effective features can be extracted from EEG signal by using the methods with variance and standard deviation.The nonuniform time window method had better accuracy in EEG analysis.The method introduced in this paper can provide some reference value for the diagnosis of brain disease in the future.