Rationale: Deep brain stimulation (DBS) has emerged as a promising treatment for a variety of neurological disorders, especially conditions such as Parkinson’s disease, where stimulation parameter optimization can be guided by instantaneous clinical feedback. However, for disorders such as epilepsy, clinical responses can take much longer, with outcome evaluation often relying on self-reported seizure frequency, making DBS parameter optimization a prolonged and challenging process. In temporal lobe epilepsy specifically, limbic networks are targeted for stimulation and are in concert with thalamic inputs – thus it is critical to understand the thalamo-limbic stimulus-response relationship to expedite the parameter optimization process.
Methods: To investigate this, we enrolled four patients undergoing intracranial stereo-electroencephalography (sEEG) monitoring for seizure localization with coverage of the anterior nucleus of thalamus (ANT). Repeated single electrical pulse stimulation (RSEPS) was delivered through the ANT contacts. Two subjects were stimulated under five charge density levels (5, 10, 15, 20, and 25 μC/cm
2/phase) and four pulse widths (50, 100, 200, and 300 μs), whereas the remaining two subjects received only two lowest charge density (5 and 10 μC/cm
2/phase) at the same pulse widths. The selected parameter space encompasses stimulation parameters proven to be safe and commonly utilized clinically. We first confirm the engagement of limbic networks and then systematically measure the response of key limbic structures to varying stimulation parameters. We quantify the stimulation effects with peak low-frequency (1-30 Hz) power increment during 10-300 ms poststimulation.
Results:
Our results showed a significant stimulation response within the limbic network (p< 0.05), including the hippocampus, amygdala, and entorhinal cortex, aligning with established anatomical findings. Generalized linear mixed-effects models reveal charge density as the primary determinant of response strength (β=0.1248, p< 0.0001), though shorter pulse widths enhance responses at matched charge (β=0.0043, p< 0.0001). Although the parameters’ influence on the response is generalizable across patients, the response strengths are patient-specific.To address this, we used Gaussian process regression (GPR) to build an individualized continuous stimulus-response map. For one patient, with only 8 sampling points used as GPR training data, the model predicts the remaining responses with an R2=0.68 ± 0.22.
Conclusions:
This study systematically characterizes ANT-limbic stimulus-response patterns across a group-level parameter space. Importantly, we developed an infrastructure for constructing personalized response maps even with sparse sampling. These findings offer a potential pathway to accelerate and personalize DBS parameter optimization for epilepsy.
Funding: This work was supported by NIH grant UH3 NS119834.