Authors :
Jiahao Chen, B.S. – University of Pittsburgh Medical Center
Presenting Author: Adway Gopakumar, BS – University of Pittsburgh
Jorge González-Martínez, MD,PhD – University of Pittsburgh Medical School
Rationale: The Virtual Brain (TVB) initiative converts multimodal clinical data into large-scale brain-network simulations and has demonstrated promise for mechanistic insight and treatment planning in epilepsy [1–4]. In collaboration with the EBRAINS/TVB team, we implemented the first U.S. point-and-click interface that harnesses high-performance computing and Bayesian optimization to create a patient-specific “virtual twin.” This twin permits in-silico delineation of the epileptogenic zone (EZ) and rapid testing of surgical or neuromodulatory strategies before entering the operating theatre. We seek to evaluate the practical feasibility of integrating TVB, advanced imaging, and stereoelectroencephalography (SEEG) into a single, clinician-friendly platform for individualized management of drug-resistant epilepsy (DRE).
Methods:
For each patient, SEEG recordings were fused with T1-weighted MRI, diffusion MRI, and post-implant CT. Following MRI–CT co-registration, subject-specific structural connectomes were generated via probabilistic tractography and parcellated with user-selected atlases. A five-dimensional Epileptor neural-mass model was instantiated at every node, with a gain matrix mapping simulated source activity to recorded SEEG channels. Bayesian inference iteratively tuned model parameters to minimize the error between simulated and empirical signals. All image processing, connectome construction, source inversion, and parameter optimization were executed from a single graphical user interface on a modern institutional workstation.
Results:
The automated pipeline consistently produced a Virtual Epileptic Patient (VEP) comprising ~260,000 vertices (≈1 mm² resolution) in less than four hours of wall-clock time. Simulated single-pulse stimulations accurately reproduced SEEG responses. In our first clinical case, the model identified three candidate EZ sites; all three matched the multidisciplinary post-hoc consensus and corresponded to regions previously resected, supporting the face validity of the approach.
Conclusions:
Our early experience shows that a fully integrated VEP interface can transform routinely acquired imaging and SEEG data into patient-specific digital twins that localize the EZ and provide in-silico guidance for precision epilepsy surgery and neuromodulation. While the present pilot is limited by sample size, it demonstrates day-to-day clinical feasibility and lays the groundwork for prospective validation in larger cohorts—an essential step toward routine adoption of virtual-brain technology in the care of patients with DRE.
Funding: None.