Detection of Focal and Non-focal Epileptic Seizure Using Continuous Wavelet Transform-Based Scalogram Images and Pre-trained Deep Neural Networks

Abstract Epilepsy is a neurological disease from which a large number of younger and older people suffer all over the world. The status of the patients is primarily examined by using Electroencephalogram (EEG) signals. The most important part for successful surgery is to locate the epileptic seizure in the brain. For this reason, it is very useful to detect the seizure area automatically before surgery. In this research, a novel method based on continuous wavelet transform (CWT) and two-dimensional (2D) convolutional neural networks (CNNs) has been proposed to predict focal and non-focal epileptic seizure. The AlexNet, InceptionV3, Inception-ResNetV2, ResNet50 and VGG16 pre-trained models have been used to automatically classify 2D-scalogram images into focal and non-focal epileptic seizure. The performances of 5 pre-trained models were compared and the detection results of 2D-scalograms were examined. The best classification accuracy of 92.27% is yielded by the InceptionV3 model among the other used four pre-trained models. As a result, it may be said that the pre-trained models and 2D-scalogram images of focal and non-focal EEG signals will be useful to neurologists for rapid and robust prediction epileptic seizure before surgery.