Abstracts

Rationale: Patients with drug-resistant epilepsy often undergo stereotactic electroencephalography (SEEG) to facilitate the localization of epileptogenic regions before surgery. However, this process can be uncomfortable for patients, as it involves triggering seizures through methods such as medication withdrawal. Furthermore, patients may not have enough seizures during SEEG monitoring to confidently predict which regions should be treated surgically. In this work, we sought to determine if network analysis of resting-state SEEG data could localize epileptogenic zones. Methods: Alpha band (8-12 Hz) imaginary coherence and partial directed coherence between SEEG electrode contacts were calculated to measure functional connectivity from a two-minute resting-state data segment in 25 focal epilepsy patients. For each electrode contact, we calculated average connectivity to contacts within the same region, and to contacts across all other regions. We then used the Brain Connectivity Toolbox to calculate eigenvector centrality, nodal eccentricity, nodal betweenness centrality (NBC), in-strength, out-strength, within module degree z-score (WMDZ), and participation coefficient.1 Measures that were significantly different between contacts in epileptogenic and non-epileptogenic regions were used to train a logistic regression model to classify epileptogenicity of regions. Results: Imaginary coherence was higher in epileptogenic versus non-epileptogenic regions (paired t-test, p<0.05, n=25). Furthermore, electrode contacts in epileptogenic regions had different within connectivity, across connectivity, WMDZ, participation coefficient, eigenvector centrality, NBC, and nodal eccentricity values compared to those in non-epileptogenic regions (paired t-test, p<0.05, n=25, corrected). In all patients, using a model incorporating these measures, we were able to classify a region as epileptogenic with a sensitivity of 71.8% and specificity of 67.3%. In a subset of patients who were seizure-free post-resection or had seizure reduction after responsive neurostimulation, only WMDZ, NBC, eigenvector centrality, nodal eccentricity, and within connectivity differed between contacts in epileptogenic and non-epileptogenic regions (paired t-test, p<0.05, n = 16, corrected). In this subset of patients, the model incorporating these network measures had a sensitivity of 89.8% and specificity of 56.0% for classifying epileptogenic versus non-epileptogenic regions. Conclusions: In this work, we found that a model incorporating network analyses of two minutes of resting-state SEEG data may aid the localization of epileptogenic regions. Further developed network analysis based models may assist clinical assessment of epileptogenic regions to target with epilepsy surgery.References: 1. Rubinov M, Sporns O. Complex network measures of brain connectivity: Uses and interpretations. Neuroimage 52, 1059-69 (2010).  Funding: This work was supported by NIH R00 NS097618 (DJE), R01 NS095291 (BMD), R01 NS0757270 (VLM), R01 NS110130 (VLM), T32 EB021937 (HFJG), T32 GM07347 (HFJG), F31 NS106735 (HFJG), and the Vanderbilt Institute for Surgery and Engineering (VISE).