Localization of epileptic focus using SVM and PSVM

Localization of epileptic focus and motor regions, is presented in the paper using raw referential EEG data from database http://eeg.pl/epi of Warsaw Memorial Child Hospital. The study is carried out on two patients named CHIMIC and JANPRZ who were diagnosed with drug-resistant epilepsy and were subsequently operated. Along with EEG recordings, inter-ictal discharges, magnetic resonance imaging (MRI) scans, structural placement of the epileptogenic zone and clinical description, are used in the study. The key step involved in the analysis, is to create an algorithm that shapes the problem into an appropriate machine learning framework, identifying the features, forward modeling of the extracted features and finally classifying them as epileptic seizures arising from the location on the right or left region of the brain. The feature extraction is done by independent component analysis (ICA) with spatially filtered eight spectral features per components. These features are classified using Support Vector Machines with radial basis function kernels (SVM) and Proximal SVM. A comparative study is done for evaluating specificity and sensitivity using segments of ICA data. The analysis and results of both patients indicates that the specificity and sensitivity ratios are higher for ICA with SVM rather than those obtained from PSVM. Localization of epileptic focuses from the classified results showed that they matched with the marked epileptic regions in the MRI scans.