Multi-model decision-making seizure types classification based on transfer learning

Abstract

For brain doctors, knowing the type of seizures in patients with epilepsy is conducive to accurate treatment of patients. Although many researchers have done enough research on the prediction and detection of epileptic seizures, there is a lack of research on the classification of seizure types. Therefore, in order to classify the 8 types of epileptic seizures, this article proposes two ideas based on the fusion of multi-models of transfer learning. Extract the MAS features from the 22 montage combined channels, use reliefF for feature selection, and finally convert it to images. Use the following two transfer learning strategies:(1) Transfer learning multi-model feature fusion, (2) Transfer learning multi-model classification probability fusion. On the TUSZ data set of temple university school of medicine, use six pre-trained models of Alexnet, Googlenet, Inception-v3, Resnet18, Vgg16, and Vgg19. The results show that the proposed algorithm are better than the compared algorithms. The probability fusion strategy is adopted to obtain the best classification performance, classification accuracy and F1 score reaching 98.48% and 97.61 %, respectively.