Detection of Epileptic Seizure for Data Augmented EEG Signals Using the Deep Learning Approaches

Abstract

Researchers working in medical imaging and signal processing are faced with difficulties due to the wide range of possible tumour forms, locations, and presentation strengths, as well as the seizures occurrence in peoples. The proposed work focusses on to identify the presence of Epileptic Seizure in Electroencephalogram (EEG) using data augmentation methods, to distinguish between the existence of stress and anxiety based Epileptic Seizures in EEG signals by hyperparameter tuned Particle Swarm Optimization (PSO) using Long Short-Term Memory (LSTM) classifier. newlineThe presence of EEG Epileptic Seizure signal has been augmented using data augmentation methods such as Position Data Augmentation (PDA) and Random Data Augmentation (RDA) followed by fuzzy based statistical feature extraction. The proposed methods such as (i) FCM-PSO-LSTM (ii) PSO-LSTM have been performed to actuate the network method in addition to retrieve seizure presence for data augmented signals. The proposed Deep Learning approaches have been verified and assessed an actual EEG Epileptic Seizure signal and Non-Epileptic Seizure signals. The novelty in proposed work categorises whether the seizure is appears or not for accuracy of (i) FCM-PSO-LSTM is about 98.5% and 98% for APDA (After Position data augmentation), and ARDA (After Random data augmentation) respectively and (ii) PSO-LSTM of about 97% and 98.5% for APDA and ARDA respectively newline