Abstracts

Rationale: Patients with refractory temporal lobe epilepsy (rTLE) are evaluated for surgical resection. Delineating the precise resection necessary for elimination of future seizures remains a significant challenge. Visual analysis of EEG patterns enables identification of probable seizure onset zones (SOZ), but does not ensure that resection of the identified SOZ will lead to seizure freedom. Improved characterization of ictal networks may improve patient selection and likelihood of seizure freedom. Graph theoretical methods provide attractive tools for network characterization.Methods: We evaluated 117 seizures from 13 subjects who underwent standard presurgical evaluation and anterior temporal lobectomy for rTLE with at least one year of followup. Intracranial EEG included bilateral frontal and temporal strips and hippocampal depths in all patients and was sampled at 200 per second. Each EEG file included a seizure, a period of minutes preceding the seizure, and up to 2 minutes following seizure termination. EEG data was processed using MatLab 2009b (Natick, MA) and the brain-connectivity toolbox (Rubinov and Sporns 2010 NeuroImage 52:1059-69). The Synchrony Index (SI), which combines local coherence and long distance phase locking, was generated over one second time bins for every pair of electrodes for all EEG files. Graph theoretical measures of network properties, including degree, strength, clustering coefficient (CC), and betweenness centrality (BC), were generated using electrodes as nodes and SI values as weighted, non-directional edges. For all EEG files, these measures were displayed for all electrodes versus time. Average electrode versus time displays were generated for each subject across seizures (Fig 1). Mean graph theoretical node measures were calculated for each hemisphere over 40 seconds spanning ictal onset and compared across ILAE outcomes using ANOVA.Results: During seizures, there were both focal and diffuse network changes in degree, strength and BC. Focal changes tended to occur at the SOZ (Fig 1). At ictal onset, both mean BC and node strength were higher in the hemisphere of SOZ than the contralateral hemisphere (p<0.01 and p=0.04). There were trends for lower mean node strength and degree among subjects who became seizure free (p=0.07 for both). A trend was seen towards higher CC in the hemisphere of SOZ (p=0.07). BC showed a trend toward interaction between hemisphere and ILAE outcome, with lower values in the hemisphere contralateral to SOZ in patients who became seizure free (p=0.15) (Fig 2).Conclusions: Node degree, strength and BC highlighted the SOZ, and each displayed consistent spatiotemporal topography within subjects across seizures. Mean BC and node strength differed by hemisphere, suggesting they may represent neurophysiological differences that lateralize the SOZ. Trends towards differences among ILAE groups in BC and node strength suggest that these measures may predict surgical outcome among patients whose EEG patterns are indistinguishable by standard visual analysis.