Detection Of High Frequency Oscillations In Accurate Location Of Seizure Onset Zone

Epilepsy is a worldwide influence on a wide range of neurological diseases.It is a kind of chronic brain dysfunction syndrome,which is caused by the abnormal discharge of neurons in the brain.Traditional seizure onset zone(SOZ)location is by the doctor according to long-range Intracranial Electroencephalogram(IEEG)machine,for patients after the EEG acquisition through visual artificial marking judgment,this way is more attention to the frequency of less than 80 Hz.Routine EEG signals acquisition time is long,greatly increases the patient's pain.Surgery results easily influenced by doctor's subjective factors or other objective factors.In recent years,more and more scholars have begun to study high frequency oscillations(HFOs),which is more than 80 Hz,and its automatic detection method has become an increasingly hot topic for the preoperative localization of SOZ.Firstly,this paper studies the preprocessing and feature extraction methods of EEG data.The data is preprocessed using data normalization,Chebyshev IIR bandpass filtering and 50 Hz frequency doubling power frequency notch.The Teager energy operator,power spectral density and wavelet entropy are used to extracting HFOs features.Through data preprocessing and feature extraction,it lays the foundation for the analysis of data and features below.Secondly,this paper proposes three methods for detecting and analyzing the HFOs of EEG: Accurate localization of SOZs based on multi-feature extraction and wavelet time-frequency map.This method analyzes the results of the three feature extraction methods of Teager energy operator,power spectral density and wavelet entropy to detect the channel of suspected SOZ,and locates HFOs in time by using the wavelet time-frequency diagram for the suspected channel.And then judge the channel of SOZ;A fast method for detecting and analyzing HFOs from classification to clustering.This method initially screens the channel of normal EEG and suspected HFOs by factoring machine.This process greatly speeds up the detection of channels.The fuzzy C-means clustering is used to analyze the components of suspected HFOs,and then make an accurate judgment;A method for detection HFOs from two classification to multiple classification.This method is aimed at the characteristics of less labeled samples and more unlabeled samples in EEG processing.Firstly,the clustered center of the semi-supervised K-means algorithm is initialized by using labeled samples,and then the EEG data of the patient is initialized by using the initialized cluster center.Cluster analysis by Mean Shift is performed for the channel of a large number of HFOs.This method does not need to set the number of clusters in advance,which greatly reduces the steps of the method and improves the calculation speed.Finally,the EEG data of 5 patients were tested by the three methods proposed in this paper,and the results were compared with the results of the three commonly used algorithms.By comparison,it can be found that the three methods proposed in this paper have greatly improved both in terms of detection time and the accuracy of localization.