Abstracts

Rationale: Functional connectivity resting state functional MRI (rsfMRI) has been used to investigate language networks in patients with epilepsy with potential to support pre-surgical planning. It is unclear how varying seed-based rsfMRI methods affect language network functional connectivity patterns and interpretation, with implications for pre-surgical planning.   Methods: We compared patterns of resting state language functional connectivity through seed-based correlation analysis at whole brain level over 12 regions of interest (ROIs) for 34 patients who underwent pre-surgical evaluation for drug-resistant epilepsy (mean age, 33 years; SD 12; 12 female).  Each patient had 3 rsfMRI runs collected in the same session with 2 conditions: fixation on a central crossbar (1 run) and eyes closed (2 runs). Seed ROIs were generated from language task (Auditory Descriptive Decision Task and Auditory Categories) peak activations at individual and group level as well as cortical atlas regions traditionally associated with language, inferior frontal gyrus (Broca’s area) and superior temporal gyrus (Wernicke’s area).  Spatial overlap between functional connectivity maps from each pair of seed ROIs was calculated as the Dice Similarity Coefficient (DSC) at individual level across all runs. ROIs generated from task activation versus atlas regions were compared, as well as within task and atlas ROIs. Spatial overlap was calculated at varying thresholds to examine relationships between overlap and threshold. We examined language network reproducibility through pairwise correlations of seed ROI time series for each run and condition, as well as spatial overlap of language networks within patients for single ROIs across 3 runs.  Results: At statistically significant thresholds (correlation coefficient r=0.20, p < 0.05), language networks had substantial spatial overlap within atlas generated ROIs (mean DSC > 0.60), fair spatial overlap within task generated ROIs(mean DSC > 0.20), and moderate spatial overlap between task and atlas ROIs (mean DSC > 0.40) . 78% of patients had moderate to substantial overlap for atlas generated ROIs, 27% for task ROIs, 42% for between task and atlas ROIs. Pairwise correlations revealed reproducible patterns across run conditions (fixation on a crossbar vs. eyes closed) and within repetition of eyes closed for the relationship between all ROI seed time series. As the correlation threshold increased, the spatial overlap between language networks decreased. Conclusions: These findings suggest resting state language networks are moderately reliable across seed method and run condition within individual patients with epilepsy. When task information is not available, functional connectivity from atlas ROIs may be sufficient to identify language networks. Resting state fMRI is a potential pre-surgical planning tool for patients with drug-resistant focal epilepsy.  Funding: NINDS NIH  Division of Intramural Research.