Epilepsy EEG Classification And Seizure Prediction Based On Machine Learning

Classification studies of epileptic EEG can help clinicians quickly identify epileptic disorders,thus freeing them from heavy workloads and improving diagnostic efficiency;prediction studies of epileptic EEG can allow patients to prevent seizures through therapeutic interventions,greatly reducing the probability of accidental injury to patients.Classification and prediction studies of epileptic EEG have far-reaching implications for improving population quality and health level.Machine learning algorithms have been widely used in various fields in the last decade or so,especially in pattern recognition and prediction studies with great results.In this paper,various machine learning algorithms are used to classify and predict epileptic EEG signals,and the specific work is as follows.Firstly,the epileptic EEG signals were preprocessed.The data were collected from CHB-MIT database,University of Bonn EEG database and Neurology database of New Delhi.Wavelet transform was used for denoising,and spectral asymmetry analysis was used to obtain the optimal single channel of patients,and follow-up research was carried out through the optimal channel data.Secondly,Inception Net-V3 convolutional neural network,SVM,K-NN,and logistic regression were used to classify the epileptic EEG,time-frequency analysis and edge-centric networks feature extraction were used.The EEG signals were converted into time-frequency maps,and normal EEG and seizure-phase EEG were distinguished using Inception Net-V3 convolutional neural network.Among the 24 patients in the CHB-MIT database,the highest classification accuracy was 95.8%,and the accuracy rates of the University of Bonn EEG database and Neurology database of New Delhi were 90.2% and 96.2%,respectively;meanwhile,used edge-centric network feature extraction method,SVM,K-NN,and logistic regression were used for classification,the highest classification accuracy was 96.9% in the CHB-MIT database,99.3% in University of Bonn EEG database and 98.6% in Neurology database of New Delhi.The results indicate that edge-centric network feature extraction can achieve more desirable classification results.Finally,time-frequency analysis and edge-centric network feature extraction were used to predict epileptic seizures.The data of 20 minutes before seizure were analyzed,and the data length of 5 minutes was divided into four time periods of 0-5 minutes,5-10 minutes,10-15 minutes and 15-20 minutes before seizure,and the data of the four time periods were predicted,and it was found that the prediction accuracy in 0-15 minutes before seizure was higher than the prediction accuracy in the time period of15-20 minutes.The results showed that a better prediction could be achieved within the first 15 minutes of seizure.In this paper,time-frequency analysis and edge-centric network were used.When classifying epileptic EEG,the feature extraction method of edge-centric network could achieve better classification results;when predicting seizures,the differences between the two feature extraction methods were not obvious,but the results were consistent.The results show that the method in this paper can be used for epilepsy detection and seizure prediction research,it is of great significance to the automatic detection and prediction of epilepsy.The feature extraction method of edge-centric network provides a new idea for epilepsy detection research,and it is important for understanding the pathological symptoms of the brain and clinical diagnosis.