An Approach of Nonlinear Energies over EEG Rhythms for Epileptic Seizure Detection
DOI: 10.66838/ma.241
Rafik Djemili LRES Lab., Université 20 Août 1955-Skikda, Algeria.
Mohammed Salem Department of Electrical and Electronic Engineering, Sabahattin Zaim University, Istanbul, Turkey.
Lamia Benzaid Department of Mechanical Engineering, Université 20 Août 1955-Skikda, Algeria.
|
Abstract Detecting epileptic seizures from electroencephalographic (EEG) recordings remains a central task in the diagnosis of epilepsy and in the localisation of the epileptogenic zone. In this work we study how well nonlinear energy measures, evaluated separately on each EEG rhythm, can discriminate between seizure and non-seizure activity. Every EEG segment is split into its delta, theta, alpha, beta and gamma components by three alternative tools, namely a bank of Butterworth (IIR) band-pass filters, the discrete wavelet transform (DWT) and wavelet packet decomposition (WPD). On each component we compute five quantities (exponential energy, Teager energy, log energy, log-energy entropy and an energy derived from the bottleneck of an autoencoder), and the resulting 25 values form the input of three classifiers: a multilayer perceptron (MLP), a support vector machine (SVM) and a long short-term memory (LSTM) network. Perfect accuracy (100%) is reached on the Bonn recordings for both the normal/ictal and the interictal/ictal problems, while 99.9% and 100% are obtained on the Single Electrode and Bern-Barcelona collections. These figures are at least as good as those of several recently published approaches.
Keywords
Electroencephalography (EEG), epileptic seizure detection, nonlinear energy, Teager energy, wavelet packet decomposition, discrete wavelet transform, neural networks., |