Localization Of Epileptic Seizure Onset Zone Based On High Frequency Oscillations Signal Detection

Epilepsy is a kind of brain disease caused by the abnormal discharge of neurons which results in the temporary disorder of the central nervous system.The form and intensity of seizure are complex.Besides,about 30%of patients with epilepsy do not get relief after taking drugs.The hope of them is to have a surgery to remove the epileptic zone.The seizure onset zone was a significant marker for the localization of epileptic zone during preoperative evaluation,and its localization accuracy determined the therapeutic effect of surgery.Therefore,how to locate accurately the seizure onset zone is the most important research significance.At present,many studies have shown that there are certain differences in the high frequency oscillations signals between the epileptic seizure onset zone and the outside,which can be used to locate the seizure onset zone.However,the traditional high frequency oscillations signal detection mainly relies on manual identification,which is inefficient and highly subjective.Therefore,an automatic detection model of high frequency oscillations signals was designed in this study.On this basis,the high frequency oscillations of each electrode contact was combined with the low frequency electroencephalo-gram characteristics to jointly locate the seizure onset zone.The research content of this topic mainly includes the following three partsFirstly,aiming at the problem that there is no public data set for high frequency oscillations signals,this paper proposes a preliminary screening algorithm to improve the calibration efficiency of signals.The algorithm establishes a threshold based on the amplitude envelope of the signal and automatically selects the suspected fragments.The people only needs to make the final determination of the suspected fragments.A total of 2700 cases of high frequency oscillations signals have been marked until now.Secondly,this paper designs an automatic detection model of high frequency oscillations signal.Wavelet packet transform is used to decompose the signals to extract the features of each frequency band,and then the improved feature selection algorithm is used to screen the features.Finally,the real and false high frequency oscillations signals are classified by support vector machine.The experimental results show that the improved feature selection algorithm is helpful to improve the detection effect,with the detection sensitivity of 89.7%and specificity of 90.7%.Finally,two kinds of seizure onset zone location algorithm were invented in our work.(1)The occurrence rate and average energy charact-eristics of high frequency oscillations signals in the electrode contacts were selected,and the seizure onset zone was located by the threshold method.The classification sensitivity of this method was 78.2%,and the mis-detection rate was 14.2%.(2)The method of feature extraction was improved.The characteristics of high frequency oscillations and low frequency electroencephalogramin the electrode contacts were combined,and the location of the seizure onset zone was realized by Decision tree,K-nearest neighbor and support vector machine algorithm.The experimental results show that the performance indexes of the improved algorithm are improved,with sensitivity of 81.2% and specificity of 83.6%.