Abstracts

Rationale: When evaluating a patient for surgical treatment of epilepsy it is critical to identify cortical regions necessary for core functions such as language that must be preserved. The gold standard for identifying such areas is direct cortical electrical stimulation (DCES). DCES is effective but is an invasive, time consuming procedure that cannot always be performed. A potential alternative to DCES is to segregate regions according to their resting state functional connectivity (RSFC). It is well established that grouping together cortical regions with similar RSFC, as measured by functional magnetic resonance imaging (fMRI), can identify areas involved in sensory, motor, and cognitive functions. However, it is not yet clear if RSFC can identify functional regions with sufficient precision in a single individual to be clinically useful, or if measures of RSFC besides fMRI are effective for defining functional areas. Here we evaluate the ability of RSFC as measured by high gamma band power (HGBP) in the electrocorticogram (ECoG) and fMRI to define functionally distinct regions via cluster analysis.Methods: HGBP fluctuations between pairs of ECoG electrodes used for DCES were derived from 5-7 minutes of data while participants were awake and immobile. The correlation between HGBP was used to quantify RSFC. These pairs of electrodes were then clustered together according to either a single or average linkage rule. The number of clusters was determined as the number that best approximated the continuous correlations. An analogous procedure was applied to 5-9 minutes of resting-state fMRI data and, as a control analysis, to the distance between electrodes. DCES was used to define 12 types of areas (e.g., hand sensorimotor, expressive language) and to evaluate the accuracy of clusters.Results: We applied the clustering procedure to data from eight individuals undergoing evaluation for surgical treatment of epilepsy. Mean compatibility (Dice s coefficient) with ESM results for HGBP was around 0.6 but only around 0.4 for fMRI, a significant difference (p<.02). This degree of agreement is better than that expected by chance (p<0.0004). HGBP but not fMRI agreement was better than when clustering on electrode distances (p<.003 & p>.42). The degree of cluster-ESM agreement did not differ between the two clustering algorithms (p=.65).Conclusions: Our results demonstrate that there is some utility of ECoG-RSFC for identifying functional regions that could complement DCES, though the agreement between the two methods is currently moderate. fMRI-RSFC was less effective and was not clearly more informative than the distance between electrodes. We suspect that fMRI-RSFC performance could be improved with alternative preprocessing (e.g., better motion correction) and by including more of the brain in the cluster analysis. The fact that both average and single linkage clustering algorithms performed equally well suggests that DCES results form neither clearly clumped nor elongated clusters.