Abstracts

Rationale: We propose a novel method for the characterization of synchronization and coupling effects in multichannel EEG and ECoG. Our method allows for visualization of the spatio-temporal evolution of synchronization effects which are characteristic of epileptic seizures, in particular for recordings with a high number of channels. The method can assist the examination of epileptic seizure propagation that is required, e.g., in the pre-surgical evaluation in epilepsy monitoring units. An important feature of our algorithm is that in contrast to numerous existing methods no channel-pre-selection is required in order to obtain stable results. Methods: The proposed method is based on a linear spatio-temporal regression analysis that is performed for each output signal separately. A channel selection algorithm determines an optimal spatial neighbourhood for each channel before the regression is computed. In this regression the samples of their own past are slightly penalized compared to the neighbourhood channels in order to emphasize the synchronization and coupling effects. Based on variances we introduce a novel measure termed extrinsic-to-intrinsic-power-ratio (EIPR), which is physiologically meaningful and valuable. The visualization of the spatio-temporal evolution of this measure allows for tracking the propagation of synchronization and coupling effects during epileptic seizures. Results: In our experiments, we used ECoG recordings from patients suffering from temporal lobe epilepsy, which were obtained during pre-surgical evaluation. The number of channels was between 28 and 32. For one patient, the proposed method was applied and the results were visualized on temporally equidistant topographic maps of the electrode positions (see fig. 1). The level of synchronization and coupling effects between channel pairs was indicated by the thickness of arrows between the respective electrode positions. Arrows below a manually chosen threshold were removed in order to obtain a clear picture. The resulting maps for the evolution of epileptic activity are in excellent agreement with the description of clinicians who visually inspected the raw ECoG signals independently of our analysis. The enclosed figure shows one example, where, according to clinicians, the seizure starts at 12:45:45 in the right hemisphere of the brain, spreads out to the left at 12:46:04, then reduces its activity and finally ends in the right hemisphere at 12:47:00. Conclusions: Comparing the results of our proposed method with the descriptions from clinical experts shows that our method can be utilized for computer-assisted evaluation of EEG/ECoG recordings of epileptic seizures. Without any manual channel or data pre-selection we obtained results which are in excellent agreement with the findings of clinical experts. The method can be applied to identify synchronization patterns even in the case of high numbers of channels, since the set of synchronized channels is determined by an automatic channel-selection algorithm. This is a particularly important feature for the practical usefulness of the method in clinics.