Research On Location Of Epileptic Seizure Onset Zone Based On Time-Frequency Analysis And High-Frequency Oscillations Detection

With the rapid development of artificial intelligence and biomedical engineering,machine learning technology has gradually become an important tool for doctors to diagnose and analyze diseases in the medical field.Epilepsy is a common neurological disease with a large number of clinical patients.For patients with medically refractory epilepsy,the success rate of neurosurgery depends largely on the accurate location of the region of onset of epilepsy.In the preoperative evaluation,the high-frequency oscillations in the epileptic electroencephalogram(EEG)signal of epilepsy has become an important electrophysiological marker to locate the seizure onset zone.Due to the weak and transient nature of high-frequency oscillations,and the mixed artifacts and noises in epileptic EEG,the traditional high-frequency oscillations detection method based on doctors’ visual judgment is time-consuming and highly subjective.In this paper,based on time-frequency analysis,the automatic detection method of high-frequency oscillations is studied,and the accurate location of the seizure onset zone is realized on this basis.According to the characteristics of epileptic EEG data,this paper studies the time-frequency analysis techniques such as wavelet transform and unsupervised clustering algorithms of signals.In this paper,a high-frequency oscillations detection model based on gaussian mixture clustering algorithm is established by using unlabeled multi-lead epileptic EEG data.Based on the experience of clinical experts,this model preprocesses the original EEG data and uses the sliding window to calculate the time domain information of EEG signals sectionally.After the pre-detection of high-frequency oscillations,the energy distribution differences between high-frequency oscillations and artifact signals were analyzed by Morlet wavelet time-frequency graph,and time-frequency features such as power ratio of high and low frequency bands,spectral kurtosis and spectral centroid were extracted as the input of the detection model.In addition,this paper combines K-Means++and Fuzzy C-Means(FCM)algorithms to optimize the initialization parameters of the detection model,reduce the computational complexity of the detection model,and improve the detection accuracy.After obtaining a high-frequency oscillations detection model with good performance,this paper analyzed the distribution characteristics of high-frequency oscillations detected in epileptic EEG in multiple periods such as seizure,interseizure,and sleep,and focused on the correlation between epileptic EEG in these periods,so as to design a positioning decision model for the seizure onset zone.Aiming at the difficulty of using EEG data during epileptic seizures,the model introduces quantitative indicators such as epileptogenic index,and establishes the lead classification threshold by combining the detection results of high-frequency oscillations in multiple periods,so as to accurately locate the seizure onset zone.Finally,the model proposed in this paper was validated by using the judgment of clinical professional doctors on the seizure onset zone as the evaluation standard.The model can accurately locate the seizure onset zone and assist doctors to narrow down the range of epileptogenic zone before clinical operation,which has good application value.