Abstracts

An important issue in epileptology is whether epileptic seizures can be anticipated prior to their occurrence. Of particular interest is the question whether information extracted from the EEG of epilepsy patients can be used for the prediction of seizures. Recent studies have shown a superiority of bivariate measures (which characterize the synchronization between two EEG signals recorded simultaneously from different locations in the brain) over univariate measures (derived from a single EEG signal) in distinguishing the seizure-free interval from the pre-seizure period. For a particular class of bivariate measures, namely, measures for phase synchronization, it is an unresolved issue whether the predictive performance of these measures can be improved by band-pass filtering the EEG prior to analysis. In this study we examine the influence of band-pass filtering on the predictive performance of these measures. We analyzed continuous multi-day multi-channel intracranial EEG recordings from up to now 5 patients undergoing invasive presurgical diagnostics. Recordings covered more than 600 hours and contained 25 seizures. Using two different classes of measures for phase synchronization, one based on the wavelet transform (providing intrinsic band-pass filtering) and one based on the Hilbert transform (no filtering), we calculated time profiles of these measures for the different channel combinations using a moving window technique. Wavelet filtering was performed according to the classical EEG bands and sub-bands. Using ROC statistics to discriminate between the preictal and interictal amplitude distributions of the obtained profiles, we quantified and compared the predictive performance of the synchronization measures with and without band-pass filtering. For the patients under investigation, the predictive performance of the synchronization measures using band-pass filtering did not vary significantly among the different EEG frequency bands, but reached the same values as the synchronization measures without filtering. Findings indicate that the overall predictive performance of synchronization measures of the EEG is not significantly improved by band-pass filtering of the signals. The changes in phase synchronization preceding epileptic seizures appear not to be generally confined to certain frequency bands. (Supported by the intramural research fund BONFOR of the University of Bonn and by the Deutsche Forschungsgemeinschaft.)