Detection of Epileptic Seizure Event in EEG Signals Using Variational Mode Decomposition and Mode Spectral Entropy

Abstract

This paper proposes a new automated generalized method for detecting an epileptic seizure event from the EEG signals. It consists of preprocessing, variational mode decomposition (VMD), mode sample spectral entropy extraction and thresholding rule. The preprocessing first performs the mean removal and then performs principal component analysis (PCA) based channel selection if required under multi-channel EEG recordings. The VMD is used to decompose the EEG signal into a series of bandlimited modes with specific central frequencies. The mode spectral entropies are extracted from the selected modes by discarding the baseline and artifacts based on the mode central frequencies. The extracted mode spectral entropy features are compared for detecting the presence of the epileptic seizures. The method is evaluated using the standard EEG databases such as Physionet CHB-MIT EEG database and Bonn database. The proposed method achieves an overall accuracy (ACC) of 98–100% with sensitivity (SE) of 100% and specificity (SP) of 97.5–100% for seven EEG groups of the Bonn database. Evaluation results demonstrate that the mode spectral entropy based generic method achieves an average SE = 88.55 %, SP =94.86 % and ACC = 72.92% on the CHB-MIT EEG dataset.