Epilepsy Detection Based On Independent Component Analysis Of EEG Features

Epilepsy is a clinical syndrome caused by excessive and repeated abnormal discharge of brain neurons,resulting in dysfunction of the nervous system in the brain and epileptic seizures.In the research of epilepsy detection,the usual procedures are:(1)EEG data acquisition;(2)feature extraction(and selection);(3)classification,recognition and the statistical analysis of results.At present,in the step of feature extraction,many EEG features are extracted,such as EEG features extracted based on time domain analysis method,EEG features extracted based on frequency domain or time-frequency domain analysis method,and EEG features extracted based on entropy and complexity analysis method.However,how to extract the most stable and the most separable EEG features from the extracted EEG features has not been systematically studied in classification and recognition.In this dissertation,on the premise of realizing a variety of EEG feature extraction algorithms and extracting many EEG features,the method of matrix decomposition,independent component analysis(ICA),is used to select EEG features.The ICA algorithm chosen is the Infomax ICA based on the largest information entropy.In the process of feature selection,feature selection is carried out from two aspects: the stability of data decomposition and the separability between features.The analysis results show that the ICA algorithm can effectively extract the stable and the separable EEG features from multiple EEG features.This dissertation also uses BP neural network,k-nearest neighbor algorithm(k-NN)and support vector machine(SVM)to classify and recognize the EEG features selected by the method of matrix decomposition.The analysis results show that the accuracies of classification from both BP neural network and SVM classification algorithm are all above 94%,and that from k-NN is over 88%.The classification recognition effect of SVM is the best,reaching 95.78%.It shows that the selected features based on the matrix decomposition method can be effectively applied to the classification and recognition of epileptic detection.