Investigation on Single Wavelet-Based Features of Scalp EEGs with Adaptive Thresholding Technique for Epileptic Seizure Classification

Epilepsy is a common brain disorder that is characterized by recurrent seizures. Classification, detection, and prediction of epileptic seizures are tasks that play a crucial role in epilepsy diagnosis and treatment. Wavelet-based features of scalp electroencephalogram (EEG) signals are common quantitative features that have been widely used for epileptic seizure detection and prediction. One potential wavelet-based feature is a quantitative feature obtained from logarithm of variance of wavelet coefficients of scalp EEG signals. In this work, logarithm-variance of wavelet coefficients extracted from scalp EEG data of children with intractable seizure and applied for epileptic seizure classification are investigated. An adaptive thresholding technique is applied to single logarithm-variance of wavelet coefficients of scalp EEG epochs for classifying scalp EEG epoch into either one that is associated with epileptic seizure or one that is not associated with epileptic seizure. In the computational experiments, a scalp EEG epoch is decomposed into seven levels using the 2nd order Daubechies wavelet. From the computational results, the best accuracy, sensitivity, and specificity of epileptic seizure classification achieved from the scalp EEG data of twelve children subjects are 0.9930, 0.9861, and 0.9977, respectively. Therefore, the computational results suggest that the logarithm-variance of wavelet coefficients of scalp EEG signal with an adaptive thresholding technique can be effectively applied for epileptic seizure classification.