Research On Feature Extraction And Identification Of Epilepsy Brain Signals Based On Generalized S-transform And Random Forest

Epilepsy is a kind of clinical syndrome caused by a variety of etiologies.There are about50 million people worldwide,and there are about 9 million people with epilepsy in China.The prevalence rate is about 5 ~ 7 ‰.Clinical can detect epileptic patients based on the characteristics of EEG collected online.However,due to the subjectivity and randomness of the work,the medical workers can classify the EEG only by the visual characteristics of EEG,so the evaluation and conclusion are not sufficient.Therefore,it is important to reduce subjective errors and reduce the burden of medical workers by using signal processing techniques and pattern recognition methods to detect electrical signals.In recent years,on the basis of signal processing and pattern recognition,many kinds of EEG signals processing method are proposed.However,due to the characteristics of the electrical signals and the limitations of the method,different methods will always have different problems,which can lead to the disadvantages such as noise,low recognition accuracy and less evaluation indicators in the processed brain electrical signals.In order to improve the noise interference,low accuracy and less evaluation index,etc.,this paper adopts the Butterworth band-pass filter to pretreat the noise of the Bonn University laboratory’s epilepsy brain electrical data,and then to deal with the direct current component to get pure epileptic EEG signals.Three different methods were selected to extract and classify the characteristic values of the pure epileptic brain electrical signals.Finally,the rich evaluation indexes were introduced to compare the three methods,and the evaluation results were obtained.This paper’s research work can include 3 aspects:1.In order to solve the lack of the data source and the obstacles of the inaccuracy of data source collection,this paper used the Bonn University laboratory’s epilepsy brain electrical data,this EEG data has the characteristics of discrete-time,stability,high accuracy,convenient and practical etc..Therefore,it can be satisfied of the requirements of the experimental data.2.In order to analyze the characteristic value of epileptic EEG signals and select the excellent algorithm,this paper analyzed the characteristics of epileptic EEG signals and studied the effect of wavelet transform(WT),S transformation(ST)and generalized S transformation(GST)on the eigenvalue classification of epilepsy EEG signals,then we used arandom forest algorithm in classification of epileptic EEG signals.On the basis of accuracy,we introduced the specificity,sensitivity and positive predictive value to form the four classes of evaluation index system,and optimized the evaluation results.3.In order to get the result of high classification evaluation index and perfect classification evaluation system,the experimental results of this paper showed that using the generalized S transform to extract the characteristic value can get higher classification accuracy.In this paper,according to a large number of experimental results,we obtained the average accuracy rates of classification were up to 98.564%,the average sensitivity rates of classification were up to 98.077%,the average specificity rates of classification were up to99.400%,the average positive predictive value(PPV)rates of classification were up to99.550%.Finally,this paper’s result showed that the algorithm has good classification accuracy and classification index diversification during comparing with the literature [51]~[56].The result met the expected goal.