Abstracts

Rationale: The epileptogenic region in mesial temporal lobe epilepsy (MTLE) extends beyond the hippocampus to a network of cortical and subcortical structures. The extent of epileptogenic network and clinical manifestation of seizures vary substantially across patients. Identification of patient-specific epileptogenic networks is essential for therapeutic modulation strategies and for characterizing commonly involved structures in a cohort of MTLE patients. To address this, multiple non-invasive methods including fMRI, scalp EEG, PET have been used to characterize epileptogenic networks. However, these methods either lack the spatio-temporal resolution or consistent activation to allow precise localization of epileptiform activity for characterizing epileptogenic networks. Here, we used intracranial recordings (of local field potential?), which provides drastically high spatio-temporal resolution, and thus would be an optimal method for delineating these networks. We particularly target inter-ictal spikes, rather than seizures, as the biomarker of epilepsy. This allows us to estimate the flow of information between different brain regions because inter-ictal spikes are frequent and occur over similar components of the MTLE network as seizures. Methods: Intracranial recordings of one-hour duration with frequent inter-ictal discharges were selected from pre-surgical intracranial recordings of 10 medically refractory epilepsy patients. Conditional probability and latency of propagation of inter-ictal spikes were computed using pairwise estimates of activity from all channels to generate a graph theoretical model between all nodes (channels). To estimate the influence of each node in this graph, directed node degree was defined as the total number of outward links from a node. A logistic regression model was then used to predict surgically resected channels using directed node degree, number of spikes in each channel and their interaction as predictors. The best predictor from this regression was normalized, thresholded and combined across patients to visualize a grouped MTLE network. Results: Directed node degree was the best predictor to classify the surgically resected epileptogenic zone, determined by conventional clinical methods based on ictal onsets. The grouped network map in these MTLE patients revealed strong connections between hippocampus, amygdala and regions in frontal lobe, limbic system and temporal lobes. Conclusions: When compared to imprecise non-invasive techniques and interventional, time-consuming stimulation methods, our method utilizes the routinely collected inter-ictal data to generate a map of the MTLE network. These maps allow us to a) identify targets of a broader network for neuromodulation b) identify patients having deviations from the "typical" epilepsy network who may not benefit from a resective procedure targeting medial structures and c) compare inter-ictal maps with a similar network obtained during seizure onset. Funding: This work is supported by a grant from NSF (1406556).