Decision support system for focal EEG signals using tunable-Q wavelet transform

In the present work, we have proposed an automated system to identify focal electroencephalogram (EEG) signals. The nonlinearity present in the focal (F) and non-focal (NF) EEG signals is quantified in tunable-Q wavelet transform (TQWT) framework. First, the EEG signals of both classes are decomposed into different subbands using TQWT. Different nonlinear features namely, K-nearest neighbour entropy estimator (KnnEnt), centered correntropy (CCorrEnt), and fuzzy entropy (FzEnt), bispectral entropies, permutation entropy (PmEnt), sample entropy (SmEnt), fractal dimension (FracDm) and largest Lyapunov exponent (LLE) are computed from these subbands. These features reveal the complexity present in various subbands of F and NF EEG signals. Our proposed method showed highest classification accuracy of 94.06% with least squares-support vector machine (LS-SVM) classifier using only KnnEnt features. The results of classification increased to 95.00% using three entropies (KnnEnt, CCorrEnt, and FzEnt) with LS-SVM classifier. We have obtained the highest classification performance in the classification of F and NF classes which can be used to locate the region of surgery in focal epileptic patients accurately.