Abstracts

Rationale: Based on the association between ictal onset (IO) and a neuronal network (NN), we tested the hypothesis that the ictal zone (IZ) in patients with non-lesional partial epilepsy refractory to medical treatment (PERT) originates in a single network and propagates to the contra-lateral hemisphere (inter-hemispheric theory) using Linear Pairwise Granger Causality (PGC) analysis of ictal EEG recordings.Methods: We prospectively studied ninety partial complex seizures (PCS) in twenty-three subjects in a case series study and defined seizure directionality in involved NN and compared to non-involved control NN. The EEG raw data was analyzed using linear PGC. Boot-strapping methodologies were used to address the statistical significance of the network interactions defined with PGC. Time frequency distribution was plotted for each seizure and compared versus controls. Results: There were no statistically significant differences when looking at the inter-hemispheric PGC measures obtained in subjects with seizures with an IO over the left frontal (p=0.2882), left temporal (p=0.1572), right frontal (p=0.6772) and right temporal networks (p=0.7795) when compared against a non involved NN such as the parietal network. The enclosed table summarizes the findings. Conclusions: There is no evidence of contra-lateral ictal spread in patients with non-lesional partial epilepsy intractable to medical treatment using PGC. The possible explanations of this finding are at least three-fold: 1) There is in fact no inter-hemispheric ictal propagation, 2) contra-lateral ictal propagation is not identifiable with PGC, and/or 3) The hidden element theory may apply, that is, the causal relation of two neurons can be affected by the effects of a third neuron that is not measured with PGC analysis. Although this study has several limitations, the findings will serve as a proof of principle for the development of models capable of predicting seizure directionality and for further evaluation of other established seizure prediction methodologies such as segmentation and surrogate analysis.