Please use this identifier to cite or link to this item:
Title: Validation of time-frequency and ARMA feature extraction methods in classification of mild epileptic signal patterns
Authors: Sakkalis, Vangelis
Zervakis, Michalis
Bigan, Cristin
Cassar, Tracey A.
Camilleri, Kenneth P.
Fabri, Simon G.
Micheloyannis, Sifis
Keywords: Electrophysiological aspects of epilepsy
Convulsions in children
Wavelets (Mathematics)
Issue Date: 2006
Publisher: Institute of Electrical and Electronics Engineers Inc.
Citation: Sakkalis, V., Zervakis, M., Bigan, C., Cassar, T. A., Camilleri, K. P., Fabri, S. G., & Micheloyannis, S. (2006). Validation of time-frequency and ARMA feature extraction methods in classification of mild epileptic signal patterns. International Special Topic Conference on Information Technology in Biomedicine (ITAB 2006), Ioannina. 1-6.
Abstract: Epilepsy is one of the most common brain disorders and may result in brain dysfunction and cognitive disturbances. Epileptic seizures usually begin in childhood without being accommodated by brain damage and many drugs produce no brain dysfunction. In this study cognitive function in mild epilepsy cases is evaluated where children with seizures are compared to controls i.e., children with epileptic seizures, without brain damage and under drug control. Two different cognitive tasks were designed and performed by both the epileptic and healthy children: i) a relatively difficult math task and ii) Fractal observation. Under this context, we propose two frameworks: the first is based on time-frequency analysis using the continuous wavelet transform (WT) and the Compressed Spectral Array (CSA), and the second is based on Auto-Regressive Moving Average (ARMA) modeling to evaluate the EEG signals at rest and during cognitive tasks in both groups. Initially, the analytical capabilities of the proposed feature extraction techniques were assessed in a simulated environment, and finally classification of the actual data was performed. The results suggest that time-frequency analysis methods were able to capture non-stationary activity, whereas ARMA modeling performs better on stationary signals. Classification of the actual data was successful and both approaches reached the same level of accuracy (73.4%). Higher frequency bands (beta and gamma) were apparent on frontal-parietal lobes on both math and fractal tests, while alpha band was diffused across a wider frontal network, only during the math task.
Appears in Collections:Scholarly Works - FacEngSCE

Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.