Abstracts

Rationale: Pediatric hemispherectomy candidates often have catastrophic epilepsy, which is believed to impact resting-state functional connectivity. Significant impairments frequently limit task-based fMRI, so resting-state fMRI (rsfMRI) is one emerging method to visualize lateralization and localization prior to surgery and inform the risk of postoperative functional loss. We hypothesized that rsfMRI methods could identify language and motor networks pre-surgically in children.


Methods: We identified 14 patients (8M, 6F) with fMRI as part of Phase I Epilepsy Surgical Evaluations, 11/14 had sedation during scans. Etiologies included vascular (n=6), Sturge-Weber (n=1), Encephalitis (n=5 Rasmussen’s; n=1 HSV), infections (n=1), and TBI (n=1). Most had left-sided seizure onset zones (SOZ; n=9), with 5 having right SOZ. Average age of onset was 3.19 (SD=2.64) and average age of fMRI was 7.86 (SD=4.02).

Anatomical images were acquired using a 3D T1-weighted MPRAGE sequence. rsfMRI images were obtained using 2D EPI BOLD sequences. Preoperative whole-brain rsfMRI scans were processed with fMRIprep. Motor and language networks were extracted at multiple thresholds via ICA and K-means cluster analysis, based on atlas regions of interest. Bilateral language regions were defined using Neurosynth’s meta-analysis, and motor regions were defined using a motor activation meta-analysis. We calculated the percent overlap between the rsfMRI ICA language network and the language atlas (thresholded at z = 2). This method has shown high agreement with neuroradiology’s qualitative ratings in our prior work.


Results: Language and motor networks were identified in all patients. No significant relationship was found between age at onset, age of fMRI, or use of sedation on language metrics. The average overlap with the language template was 25%, and the average Dice coefficient was 0.52. Patients with vascular etiology trended towards more atypical language networks compared to encephalitis (Vascular M Dice = 0.50, SD = 0.11; Encephalitis M Dice = 0.58, SD = 0.29). This trend was not observed for the motor network. ANOVA indicated no significant relationship between etiology and Dice coefficients (F(3, 10) = 1.173, p = 0.368), though effect size estimates suggest potential variability for future studies.

Motor networks were weaker with sedation (Average Dice without sedation: 0.69, SD: 0.17; with sedation: 0.58, SD: 0.19). Motor and language[AA1] metrics did not differ by side of epilepsy; qualitatively, language networks showed slightly less agreement with the template in left-sided epilepsy (M = 0.50, SD = 0.23) compared to right-sided epilepsy (M = 0.56, SD = 0.24). Longer duration of epilepsy was associated with lower language network agreement (Pearson R = -0.42, p = 0.138).


Conclusions: We found preliminary evidence that longer epilepsy duration may be more associated with more abnormal language networks in pediatric hemispherectomy candidates. Left-sided epilepsy shows less language template overlap, which is likely related to anatomical disruptions associated with acute or progressive disease. Future studies with larger samples should validate these preliminary findings.

Funding: R01 & Bernstein Award