Detection of epileptic seizure patterns in EEG through fragmented feature extraction

Epilepsy is a common neurological disorder hallmarked by frequent seizures. Electroencephalography plays an imperative role for diagnosis of epileptic seizures. But since, electroencephalogram (EEG) signals are contaminated with artifacts, detection of epileptic activities is a difficult task even for an experienced neurophysiologist. Therefore, development of an automatic epileptic seizure detection system is essential to improve epilepsy diagnosis. Present research proposes a novel method for analyzing EEG signals using its sections to extract features. The developed scheme involved computation of statistical features which were then passed into the least-squares support vector machine classifier with non-linear kernel. The performance of the classifier was tested with ten folds for cross-validation. The developed scheme showed mean accuracy of 97.19 % with 0.065 s computation time. Thus, this study holds promising ground for development of real-time automated epilepsy diagnosis system.