Research On Cross-frequency Coupling Features Of Epilepsy Intracranial EEG Based On Machine Learning

Epilepsy is a neurological disorder characterized by a sudden interruption of normal brain activity,caused by abnormal discharge of neurons in the brain.When an epileptic seizure occurs,various systems of the patient’s body will have functional disorders,which can threaten the patient’s life in severe cases.Intracranial EEG has become one of the most important tools for the diagnosis and treatment of refractory epilepsy due to its advantages of high signal-to-noise ratio,high spatial resolution,and high accuracy,but the massive amount of EEG data has brought heavy pressure for clinicians to analyze the data to determine the epileptogenic zone.This thesis takes epileptic intracranial EEG as the research object,analyzes the coupling features of high-frequency signal amplitude and low-frequency signal phase in intracranial EEG during epileptic seizures,and uses machine learning methods to learn and classify cross-frequency coupling features;the effect of the coupling phase feature and intensity feature on the classification performance of the machine learning method is studied,and the different types of samples are statistically analyzed from the two aspects of coupling strength and phase,and finally the location of the epileptogenic zone in patients with epilepsy is realized,the specific work is as follows:First,this thesis uses the epilepsy EEG data of 6 patients from the Northern Theater General Hospital to construct cross-frequency coupling sample features,and uses phase amplitude coupling(PAC)to characterize the high-frequency signal and low-frequency signal of the EEG signal.Based on the EEG signal fragments in each time window,the two-dimensional PAC map is calculated and constructed.Based on the PAC maps data set,convolutional neural networks(CNN)are used to train and classify the epileptogenic zone PAC map and the non-epileptogenic zone PAC map,and map the classification results to the electrode channels.The analysis results show that,based on the characteristic pattern of coupling strength,the area under the curve(AUC)for classification using the CNN method can reach 0.88.Secondly,this thesis further extracts the coupling phase information,and constructs the phase map of the coupling features.The PAC map and the phase map are used as sample features.After expanding the dimensionality of the CNN input layer,the samples are trained and classified,and the results showed that the AUC of the classification increased to 0.93,using the support vector machine method and the random forest method,based on the strength and phase information of the coupling classify the samples,and the AUC value of the classification is lower than that of the CNN method.Finally,this thesis also carried out statistical analysis on the samples of epileptogenic zone and non-epileptogenic zone.The results showed that in the θ band,the difference of the samples is mainly reflected in the coupling strength,while in the slow δ and fast δ bands,the difference of samples is mainly reflected in the coupling phase.The current results show that using the CNN method to classify the two-dimensional PAC map of the epileptogenic zone EEG signals can achieve better performance,and the addition of the coupling phase feature can help us better locate the epileptogenic zone.