Research On Epileptic Seizure Type Classification Algorithm Based On Deep Learning Multi-model Decision-making

Epilepsy is a chronic neurological disease that threatens human life and health.Due to the sudden nature of its pathology and the number of people increasing at a rate of about400,000 every year,the treatment of epilepsy is urgent.As far as doctors are concerned,it is not only necessary to predict the period of epilepsy and remind patients to take precautions in advance,but also to judge the type of epilepsy in time when an epileptic seizure occurs,so as to give accurate treatment and reduce the pain and mortality of patients during the onset.EEG is one of the most important and efficient ways to obtain brain information in epilepsy patients.In epilepsy research in recent years,most researchers focus on the prediction and detection of pre-epilepsy onset,and there is a lack of research on the classification of epileptic seizure types.If there is a research on the classification of epileptic seizure types,it will greatly shorten the treatment time,reduce the pain of patients during the onset period,and avoid death caused by untimely treatment.Therefore,research on the classification of epileptic seizures is of great significance.This research is mainly divided into the following two parts:(1)Two algorithms based on multi-model fusion based on transfer learning are proposed.On the TUSZ dataset of Temple University School of Medicine,the mean magnitude(MAS)features are extracted from 22 montage combination channels,and relief F is used for feature selection,Finally convert it to an image.The following two transfer learning strategies are used:(1)transfer learning multi-model feature fusion,(2)transfer learning multi-model classification and classification probability fusion.Use six pre-trained models of Alexnet,Googlenet,Inception-v3,Resnet18,Vgg16 and Vgg19 to replace the next three layers(full connection layer,softmax layer and output layer)for deep feature extraction,and finally feature fusion is to 6 The features of each pre-trained model are classified by a support vector machine(SVM),and the result is obtained;while the probability fusion is to classify the features of the six pre-trained models with six SVMs respectively,and finally use DS evidence theory performs probability fusion to obtain classification results.The results show that the proposed algorithm outperforms the comparison algorithms.The probabilistic fusion strategy is used to obtain the best classification performance,with the classification accuracy and F1 score reaching 98.48% and 97.61%,respectively.(2)The idea of using generative adversarial network for epileptic seizure data expansion is proposed.The multi-model fusion classification algorithm proposed in this paper is based on the sufficient and balanced data set for classification.In order to improve the generalization performance of the algorithm,it can achieve good results even when the data set is unbalanced.Using the generative adversarial network to expand the imbalanced dataset,the classification index is improved by about 3% compared with that before the data expansion.