Abstracts

Rationale: Deep brain stimulation (DBS) of the thalami has emerged as a safer and promising therapeutic option in drug-resistant epilepsies (DRE). However, the therapeutic responses are often suboptimal, with significant variability in seizure reduction observed across patients. Objective standards are needed to guide the therapy optimized to the patient’s epileptogenic network. Converging evidence from the connectomic-DBS suggests that treatment outcome is likely to be effective if the stimulation parameters are optimized to engage the patient-specific thalamocortical epileptogenic circuit. Unfortunately, in clinical epilepsy, reliable methods for target identification and engagement remain a critical knowledge gap that should be addressed to improve the outcome of this FDA-approved therapy. In this study, we show how thalamic neuromodulation can be optimized based on a patient’s thalamic nuclear segmentation, determining the integrity of its cortical white matter connectivity and, eventually volume of tissue activated (VTA) to estimate the effects of stimulation on the target nucleus and its surrounding nuclei.

Methods: A total of 19 patients undergoing deep brain stimulation (DBS) or responsive neurostimulation (RNS) (DBS: 18, RNS: 1, Anterior-ANT:9, Centromedian-CM: 5, Pulvinar-Pul: 2, CM+Pul: 2) of the thalami were included in the study. White matter nulled (WMN) MRI images were used for patient-specific segmentation using Thalamus Optimized Multi Atlas Segmentation (THOMAS) algorithm. The postoperative CT was coregistered to the preoperative T1, WMN images, and diffusion tractography scans using affine transform in advanced normalization toolbox. Generalized Q-sampling Imaging (GQI) was used to resolve fiber orientations and quantify the tracts from the thalamic nuclei to the cortex. SimBio based finite element model (FEM) was used to estimate the VTA. The clinical outcome at the latest follow-up was graded as a reduction in seizure frequency of < 50%, 50-75%, and >75%.

Results: WMN images provide good contrast to distinguish the thalamic nuclei. Distinct centromedian knob and hypodensity of the mamillo-thalamic tract can be visualized in these images. Even in cases of malformed subcortical anatomy, deep learning-based THOMAS segmentation provides a reliable definition of thalamic architecture. Distinct white tracts from various thalamic nuclei can be mapped to the cortex. In our example case (figure 1c) we show how VTA may influence not only the targeted nucleus but also the surrounding thalamic nuclei. 7/10 patients (5/6-ANT, 2/4-CM) had < 50% seizure reduction, 2/2 (CM+Pul) had 50-75% reduction, and 5/12 (1/6-ANT, 2/4-CM, 2/2-Pul) had >75% reduction in seizure frequency.

Conclusions: Our composite anatomo-physiological framework provides a future direction to optimize neuromodulation for every single patient.

Funding: None