Abstracts

Rationale: Thalamic neurostimulation through deep brain stimulation (DBS) or responsive neurostimulation (RNS) is a novel therapy for patients with refractory epilepsy. Although some encouraging outcomes have been observed, not all patients achieve sufficient seizure control. One factor that may play a role in this discrepancy is the degree of thalamic involvement in seizures. In the case of RNS, the device first needs to detect the activity spread in the thalamus to trigger stimulation. However, it is not clear which thalamic nuclei should be targeted for detection and neuromodulation in each patient. To address this, we examined presurgical intracranial recordings in patients with epilepsy to determine what seizure types are more likely to spread to the major stimulation-targeted nuclei: centromedian (CM), anterior nucleus of the thalamus (ANT), and pulvinar (PLV).

Methods: Seizures (n=717) recorded from 44 patients who had at least one electrode in a thalamic nucleus were reviewed and categorized based on their onset pattern, onset region, and spread pattern to the rest of the brain after their onset. To account for seizure count variability across patients, only five seizures of each type were selected for patients with many seizures of one type. Spread time to CM, ANT, and PLV of these seizures (n=308) was evaluated using electrographic changes in seven different measures commonly used in most commercial detection devices. The time of spread of a subset of seizures was also evaluated by an epileptologist to compare with the automated detection. Each seizure’s time of spread to the thalamus was categorized as early (within 5 s), late, or no-spread. A binomial response GLM was used to evaluate the effect of each feature on the time of spread. We used the Wald test to evaluate whether a predictor value significantly (p< 0.05) affects the time of spread.


Results: The results of our automated method were largely in agreement (70%) with the expert-identified spread time and were used for the rest of the analysis. Out of the three nuclei, PLV had the highest number of early spread (66%), followed by ANT (61%) and CM (50%). Since all three nuclei were not simultaneously recorded in all patients, GLM was used to identify if any feature could predict the spread to each nucleus. We found that seizures with hypersynchronous onset patterns are the least likely to spread to any nuclei early. For spread to ANT, seizures with mesial temporal onset that secondarily generalize are more likely to spread early to ANT, while seizures with lateral temporal onsets have an early spread to PLV. Seizures with broad onset tend to have an early spread to both CM and PLV, with PLV tending to exhibit earlier spread compared to CM, showing either can be a suitable target for these seizures.


Conclusions: These findings provide much-needed insight into identifying suitable thalamic sites for neuromodulation. Seizures have varied pathophysiology in terms of thalamic involvement, and this specificity in network involvement can be leveraged to predict which thalamic nuclei are most important to consider as targets for a particular patient based on their seizure types and characteristics.


Funding: CDMRP ERP-W81XWH-22-1-0315