Researchers Introduce Virtual Brain Twin Framework to Improve Epilepsy Diagnosis

By Soham NandiReviewed by Bethan DaviesAug 15 2025

Researchers have developed a virtual brain twin framework that accurately maps epileptogenic zones using both invasive and non-invasive stimulation, offering a promising step toward less invasive treatment for drug-resistant epilepsy.

Published in Nature Computational Science, the study introduces a modeling approach that combines individualized brain imaging with functional recordings—including EEG, SEEG, and stimulation-induced seizure data—to enhance the diagnosis of epileptogenic zone networks (EZN). By integrating both invasive (SEEG) and non-invasive (temporal interference, or TI) stimulation methods, the framework aims to support a shift toward safer, non-surgical epilepsy interventions.

Background

For patients with drug-resistant focal epilepsy, successful treatment depends on pinpointing the epileptogenic zone network (EZN)—the brain region where seizures begin. Current methods often rely on invasive SEEG electrode implantation and prolonged monitoring of spontaneous seizures, which can be risky and uncomfortable.

While recent advances like the Virtual Epileptic Patient (VEP) model have improved EZN localization using patient-specific data, they still fall short in leveraging stimulation-induced seizures and enabling non-invasive alternatives. This study addresses both gaps with a high-resolution virtual brain model designed to simulate individual brain responses to both invasive and non-invasive stimulation.

How the Framework Works

The study focused on two patients undergoing presurgical evaluation at La Timone Hospital: a 23-year-old woman with occipital lobe epilepsy and a 19-year-old man with frontal lobe epilepsy. Each patient’s MRI data, including T1-weighted and diffusion images, was processed using FreeSurfer and MRtrix to reconstruct detailed cortical surfaces and structural brain networks.

Using the VEP atlas for parcellation, researchers built individualized virtual brain models incorporating an extended version of the Epileptor model. This version included a stimulation-accumulation parameter that simulates how seizures are triggered when brain activity crosses a certain threshold.

The team simulated electric fields generated by both invasive SEEG and non-invasive TI stimulation. SEEG fields were based on the placement of implanted electrodes, while TI fields were modeled using SimNIBS software to optimize scalp electrode configurations for deep brain targeting—without surgery.

To align simulated activity with real recordings, the researchers projected model outputs onto EEG and SEEG sensors using forward models: distance-based gain matrices for SEEG and boundary element methods for EEG. They then used Hamiltonian Monte Carlo (HMC) to estimate model parameters from actual seizure data, allowing them to localize seizure onset regions based on calculated epileptogenicity values (EVs).

What They Found

The workflow began by generating patient-specific brain models containing over 20,000 vertices, enabling fine-grained simulation of seizure dynamics. The extended Epileptor model, paired with stimulation parameters, effectively replicated how seizures propagate in individual brains.

In both patients, seizures induced by SEEG and TI stimulation localized to regions consistent with their clinical diagnoses: the left occipital cortex in the occipital lobe epilepsy case and the left frontal cortex in the frontal lobe epilepsy case.

The framework demonstrated three major strengths:

1. Accurate EZN Mapping Across Stimulation Types: Model inversion using HMC consistently pinpointed epileptogenic zones that matched surgical resection areas in both cases.
2. Enhanced Precision with Multimodal Data: Combining SEEG and scalp-EEG recordings improved localization accuracy, reinforcing the value of integrated data.
3. Robustness Across Atlases: While spatial resolution varied slightly, the system maintained diagnostic performance even when switching brain parcellation schemes.

Importantly, stimulation—whether invasive or non-invasive—only triggered simulated seizures in epileptogenic regions, not in healthy tissue. This specificity is crucial for reducing false positives in clinical diagnostics.

Conclusion

This study presents a compelling virtual brain twin approach for localizing epileptogenic zones in patients with drug-resistant epilepsy. By simulating individualized brain responses to both invasive and non-invasive stimulation, the model closely mirrored surgical outcomes while improving diagnostic confidence through multimodal data integration.

The success of non-invasive TI stimulation in targeting deep brain regions signals a major step forward in reducing the need for surgical electrode implantation. The pipeline’s adaptability across brain atlases and stimulation protocols adds to its clinical value, though future work will need to address computational demands and longer-term seizure modeling.

Beyond epilepsy, this framework opens the door to personalized, simulation-based diagnostics for other neurological conditions—and potentially, to a future where non-invasive therapies are the norm rather than the exception.

Journal Reference

Wang, H.E., Dollomaja, B., Triebkorn, P. et al. Virtual brain twins for stimulation in epilepsy. Nat Comput Sci (2025). DOI:10.1038/s43588-025-00841-6. https://www.nature.com/articles/s43588-025-00841-

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.