Abstract

Epilepsy is a chronic, non-communicable brain disease and one of the most common neurological disorders worldwide. Epileptic seizures can progress from brief loss of attention and muscle twitching to severe and prolonged convulsions, and the frequency of seizures can also increase from once a year to multiple times a day, making epilepsy prediction very difficult. Moreover, due to the uncertainty, suddenness, and recurrence of epilepsy, patients often lose control of their bodies and lose consciousness, resulting in injuries and threat to life. Effective analysis and accurate classification of epilepsy can procure these episode’s and situation’s. Therefore, in this paper using deep learning models namely BiLSTM, DenseNet, and EfficientNetV2 vide epilepsy EEG dataset provided by University of Bonn’s which is based on time series subsequently, transformed into 2D dataset by using Gradient- weighted Class Activation Mapping (Grad-CAM) for seizure detection. Consequently, all three models are able to classfy Epileptic seizures whereas results show that the EfficientNetV2 model achieves the best classification accuracy for two-dimensional EEG images of epilepsy, reaching 98.69%, thus confirming the feasibility of the EfficientNetV2 model in epilepsy seizure detection.

Authors

Harshita, Pinkee, Kuldeep Tomar
NGF College of Engineering and Technology, India

Keywords

Epilepsy, Electrophysiological Signals, Deep Learning, Long Short Memory, Convolutional Neural Networks