Jacobi Polynomial Transforms-Based Entropy Measures for Focal and Non-Focal EEG Signals Discrimination Using Kernel Machines

Abstract

Electroencephalogram (EEG) remains the primary technique in the diagnosis and localization of partial epilepsy seizures. Despite the advent of modern neuroimaging techniques, the use of EEG signals for locating epilepsy-affected brain areas is still convenient. That is why during these last decades, several computer-aided detection (CAD) methodologies have been proposed to detect and discriminate focal (F) EEG signals, and hence locate epileptogenic foci. In this impetus, this paper applied Jacobi polynomial transforms (JPTs)-based entropy measures to analyze the complexity and discriminate the bivariate focal (F) and non-focal (NF) EEG signals. Jacobi polynomial transforms namely discrete Legendre transform (DLT) and discrete Chebyshev transform (DChT) are applied to separate F and NF EEG signals into their different rhythms. Furthermore, entropy measures like approximate entropy (ApEn), sample entropy (SampEn), permutation entropy (PermEn), fuzzy entropy (FuzzyEn) and increment entropy (IncrEn) are extracted. For direct discrimination between F and NF EEG signals, extracted entropies are combined to define different features vectors that are fed as inputs of two kernel machines namely the least-squares support vector machine (LS-SVM) and simple multi-layer perceptron neural network (sMLPNN). Experimental results demonstrated that our methodology achieved the highest performance of 98.33% sensitivity, 98.00% specificity, and 98.17% accuracy in discriminating F and NF EEG signals with sMLPNN classifier. In addition, our methodology will be useful to clinicians in providing an accurate and objective paradigm for locating epilepsy-affected brain areas.