Research On Prediction Of Epileptic Seizures Based On Brain Network

Epilepsy is a cerebral dysfunction syndrome second only to stroke in incidence and is caused by repeated excessive discharge of neurons in the brain.The sudden and recurrent characteristics of epilepsy cause patients to bear tremendous physical and psychological pressure.Even lead to depression and impaired cognitive function in patients,which even lead to suicidal behavior.The study pointed out that the sudden onset of epilepsy caused the patient to temporarily incapacitate the patient and lead to the dangerous situation.Such as fall and traffic accident can cause more harm than epilepsy itself.According to the survey report of the World Health Organization,there are about 65 million epilepsy patients worldwide,while the number of epilepsy patients in our country is about 9 million.About 70% of epilepsy patients can be suppressed by taking antiepileptic drugs in time.Predicting the onset of epilepsy in advance and alerting patients in time can win precious time for patients to take medication,and even suppress upcoming seizures.So it has important research significance.This paper uses the scalp EEG signals of 14 epileptic patients in three states: “interictal”,“preictal” and “ictal”,and uses the directed transfer function method to calculate the causal transmission intensity between scalp channels at different sub-band.The e Connectome toolbox was used to analyze the direction and intensity of EEG information transmission in patients in three dimensions: time,space,and frequency domain.The morphological changes of the network topology from the interictal state to the preictal and then to the seizure state were revealed in the form of network information flow.The directed transfer matrix is binarized and the effective brain network is constructed.The brain network characteristics of three states are obtained by using the method of graph theory,and the mean square deviation of network characteristics of different states is compared.Thus,it provides data support and evidence for revealing the pathogenesis of epilepsy.In the chapter of seizure prediction for epilepsy,only the EEG characteristics of "interictal" and "preictal" are distinguished.In the process of real-time monitoring of the EEG of epilepsy patients,an alarm is issued when the characteristics of the current EEG state match the characteristics before the seizure.The support vector machine method is used to train the features of the patient’s effective brain network with significant differences,so as to achieve the effect of classification.At the same time,a convolutional neural network was built to train and classify the directed transfer matrix between different brain regions of patients.Then,based on the classification model of support vector machine and convolutional neural network,two-step method is used to predict the seizures of epileptic patients.The two-step method is a two-layer joint analysis method.The first layer analyzes the EEG signals lasting for two minutes,and the second layer analyzes and determines the analysis results based on the results of the first layer lasting three times.By comparing the prediction effects of the two classification models,it is found that the prediction model using brain network features for classification has higher accuracy than the prediction model using directed transfer matrix for classification,with a prediction accuracy of 99% ?0.8%,The average prediction time was 30 minutes,and the prediction time in some patients was up to 52 minutes.