456 Connectomic Predictors of Outcome in Centromedian RNS and DBS for Generalized Epilepsy

INTRODUCTION: Although generalized epilepsy patients have traditionally not been considered surgical candidates, thalamic responsive neurostimulation (RNS) and deep brain stimulation (DBS) of the centromedian nucleus (CM) have yielded promising results. However, specific mechanisms of action and relevant brain networks are uncertain, and it therefore remains unclear how best to target these neuromodulatory interventions. METHODS: 37 patients undergoing CM neurostimulation for generalized epilepsy were aggregated, including 11 patients undergoing RNS (MGH-UPMC), 9 patients undergoing DBS (Mainz-Madrid), and 17 patients undergoing DBS (ESTEL). Patient outcomes were recorded as percent seizure frequency reduction from baseline. Volumes of tissue activation (VTAs) were generated using Lead-DBS software. Normative fiber-tracks traversing VTAs were identified using a 64K track thalamic-specific connectome generated using 32 Human Connectome Project (HCP) subjects. Fiber-track predictivity of outcome was determined using fiber-filtering methodology, with internal validation using leave-one-out methods. Top predictive fiber-tracks were then characterized anatomically. RESULTS: Normative structural connectivity significantly predicted outcomes within each cohort (MGH-UPMC p = 0.02, Mainz-Madrid p = 0.02, ESTEL p = 0.04). Normative structural connectivity was cross-predictive between DBS cohorts, but not between DBS and RNS cohorts. Top predictive fiber-tracks were significantly more likely to traverse the CM for RNS but not DBS. Top predictive fiber-tracks significantly overlapped ascending reticular activating system (ARAS) nuclei in all groups. CONCLUSIONS: Normative structural connectivity may predict outcomes of CM RNS and CM DBS for generalized epilepsy and may represent a tool for targeting. However, disparate predictive networks are implicated for CM RNS and CM DBS, suggesting the possibility of varying mechanisms of action. While the CM itself is specifically involved in the predictive network of only CM RNS, both CM RNS and CM DBS involve ARAS nuclei.