(I-342) Creating Patient-Specific Computational Models for Off-Contact Stimulation from Stereo-Electroencephalography Leads

Introduction:: Epilepsy — affecting tens of millions of people worldwide — drives episodic, uncontrollable, and irregular brain activity that manifests as spontaneous seizures. Although medications can reduce seizure severity and frequency, they fail to effectively control seizures in the roughly 30% of epilepsy patients diagnosed as medically refractory. Electrical stimulation of the typically deep-brain regions constituting the seizure initiation zone offers a possible therapy for these medically refractory patients. Stimulating devices require invasive implantation and, to be effective, surgeons must center the stimulating contacts in the seizure initiation zone. Under optimal circumstances, deep brain stimulation can reduce seizure rates by 48-98% in medically refractory patients. However, the precise location of the seizure onset zone is often unknown and imprecise contact targeting greatly reduces therapeutic efficacy. To overcome this problem, we aim to enable clinicians to steer electrical current into seizure onset zones from contacts spatially distributed throughout the brain. Such exquisite current steering requires a precise conductivity model of the patient’s brain, including the embedded contacts. We developed a semi-autonomous computational pipeline to generate realistic patient-specific brain models from patient imaging data. We tested the ability of this pipeline to enable off-contact targeting in human participants with medically refractory epilepsy.


Materials and Methods:: We assembled a computational pipeline to generate realistic, patient-specific brain models. First, pre-operative scans are overlaid with post-operative CT scans using MATLAB and a custom interface named LeGUI. This step allows the implanted leads to be visualized with high-resolution neural anatomy from the MRI scans. Next, the MRI slices are stitched together in the program SimNIBS, creating 3D surfaces representing white, gray, and ventricular matter. The resulting 3D models are used to generate a layer of cerebrospinal fluid (CSF) surrounding the brain in 3D Slicer. The surfaces, CSF layer, and lead models are then meshed together using the program SCIRun and assigned conductivities reflecting realistic values for each matter type. SCIRun is then used to perform basic voltage calculations in the brain matter between two electrodes. The results of these voltage and conductivity models are paired with a region of interest (ROI) — a hypothetical seizure onset zone — and fed to an optimization algorithm. The optimization returns electrode patterns and current amplitudes to stimulate the ROIs electrically. We tested these configurations in the human participants.


Results, Conclusions, and Discussions:: A semi-autonomous pipeline was created to generate patient-specific brain models. With minimal supervision, the pipeline returned realistic and easily visualized 3D models. The optimization algorithms generated optimal electrode configurations to stimulate specified ROIs (Fig. 1). We selected ROIs centered on existing electrode contacts and excluded those contacts from consideration in the optimization algorithm. We subsequently recorded from those excluded contacts and confirmed that the current we steered into the ROI matched the patient-specific computational predictions. Post-stimulation brain activity generally increased, showing that the models were realistic and the algorithm correctly predicted the neural response in the ROI.



Our semi-autonomous computational pipeline can generate models sufficient for an optimization algorithm to identify electrode configurations that target off-contact ROIs. These models and experiments could help improve deep brain stimulation efficacy by enabling accurate ROI targeting even after surgical implantation. In addition to epilepsy, this pipeline can assist in treating other neurological disorders, such as depression and Parkinson's disease. The semi-autonomous pipeline currently requires some user input and visual checks; future work could improve the autonomous nature of the pipeline, bringing it closer to clinical utility.


Acknowledgements (Optional): : Funding provided by the University of Utah Undergraduate Research Opportunity Program. We thank J. Herron, S. Rahimpour, J. Rolston, and T. Davis for their critical contributions to other aspects of this project.

References (Optional): :