Weighted 1D-local binary pattern features and Taylor-Henry gas solubility optimization based Deep Maxout network for discovering epileptic seizure using EEG

Abstract

Epilepsy represents chaos in nerves which can affect the world's population. Such type of abnormal activities of the brain can lead to seizures. Hence, precise and timely treatment of seizures is important to minimize financial and living costs. Electroencephalogram (EEG) is considered an imperative tool for analyzing epilepsy to diagnose epilepsy. This paper devises an optimization-aware deep model for detecting epilepsy using EEG signals. Here, the EEG signals undergo feature extraction wherein several features like relative amplitude, spectral entropy, logarithmic band power, power spectral density, Multiple kernel weighted Mel frequency cepstral coefficient (MKMFCC), tonal power ratio. The proposed weighted one-dimensional Local Binary Pattern (1D LBP) is obtained by combining weighted function in 1D-LBP are extracted. After extraction of features, data augmentation is carried out by flipping the EEG signal with the circular shift. The training of the Deep Maxout network is trained is done by the devised Taylor Henry gas solubility optimization (Taylor HGSO), by merging the Taylor Series and HGSO. The developed Taylor HGSO-based Deep Maxout network offered enhanced performance with high accuracy of 93.6%, sensitivity of 94.7%, and specificity of 93.4%.