Classification of EEG Signal Based on Pre-Trained 2D CNN Model for Epilepsy Detection

Epilepsy is a chronic neurological disease due to extreme electric discharge in the brain. Epileptic seizures gravely affect the social life and psychology of patients, knowing the lack of neurologists and the fact that analyzing EEG signals is time-consuming and must be done by an expert; therefore, it is important to correctly diagnose patients with epilepsy automatically. In this work, we start with preprocessing raw EEG data, and we extract the spectrogram of every channel. In the next step, we propose an image category classification method that combines the extractor of the feature pattern, which is AlexNet, which is a pretrained convolutional neural network (CNN), with a trainable classifier support vector machine (SVM). Alex-Net passed the feature vectors to the SVM. The spectrogram images that we extracted are used as the input layer. The number of classes is two: one for normal cases and the second for epileptic cases. We split the training and test samples. Finally, spectrogram images are trained by many classifiers, and we find that the SVM in this study had the best mean accuracy, reaching 97.52%, which means that this model is effective for epilepsy classification.