Using Neurotwins to Optimize tDCS Treatments for Epilepsy

Author: Borja Mercadal

Roughly one-third of epilepsy patients live with drug-resistant, or refractory, focal epilepsy, where seizures persist despite trying multiple anti-seizure medications. For these individuals, treatment options are limited, and invasive interventions like resective surgery or neurostimulation are often considered. However, not all patients are candidates for these procedures, either due to the location of the seizure focus or the associated risks.

Transcranial electrical stimulation (tES) has emerged as a promising non-invasive alternative, offering the potential to modulate brain activity and reduce seizure burden without surgery. But to realize this potential, especially in such a heterogeneous population, a personalized and targeted approach is essential. This process, usually referred to as montage optimization, consists of finding the optimal electrode positions and currents for each patient.

Montage optimization based on biophysical head models (tES 1.0)

A possible montage optimization strategy is based on the use of personalized biophysical head models as discussed in our Neurotwins: Personalizing Digital Brains post.  As shown in the figure below, the process begins with the structural MRI of the patient, which is used to segment various tissues. These segmentation masks are then used to build a 3D model of the patient's head geometry, which allows for simulating the electric field generated by a montage. In epilepsy, clinicians can define the regions of the patient’s brain to inhibit using Neuroelectrics' online target editor and an optimization algorithm, together with the personalized biophysical head model, to find the best montage to do so.

Neurotwins for montage optimization (tES 2.0)

In a recent post, we discussed how digital brain model twins, or Neurotwins, can be used to personalize transcranial electrical stimulation (tES) by simulating the brain’s response to the electric fields. Building on this idea, in the ERC-funded Galvani project, we have been working towards the application of these principles to optimize tDCS montage for epilepsy treatment. The goal is to go beyond simply inhibiting the epileptogenic zone (EZ) and the propagation zone (PZ) and identify additional targets, as well as avoiding excitation of sensitive areas.

How we build Neurotwins for epilepsy

In epilepsy, we use structural MRI images, diffusion MRI (dMRI) images, and stereo-Electroencephalographic (SEEG) recordings of the patients to build a Neurotwin. The whole process is summarized in the image below:

The MRI images are used to create a biophysical head model following the same process as in tES 1.0. The dMRI images are used to obtain the connectome, which is a map of how strongly the different brain regions are connected to each other. The SEEG recordings allow the clinicians to identify the EZ and PZ regions.

With this information, we can begin building a personalized brain network model (BNM). To do so, we use interconnected neural mass models. In the regions identified as EZ, a novel mathematical model that generates seizures developed by our time, while in the rest of the brain, we use a model inspired by the work from Wendling et al, which is capable of propagating seizures but not of entering into seizure by itself.

We set the connections between these regions following the connectome, and we tune the model’s parameters to reproduce the early seizure spread seen in the patient’s SEEG recording. This process is done with the help of an evolutionary algorithm, and once it is finished, we have our patient’s Neurotwin ready. If you want to know more about how we create Neurotwins for patients with epilepsy, check out our new paper on Personalized Whole-Brain Models of Seizure Propagation.

Neurotwin-based montage optimization

Once the Neurotwin is in place, we can begin the search for an optimal electrode montage. The process to evaluate a montage begins by analyzing its effects on seizure spread as follows:

1. We estimate the electric field generated by the montage using the personalized biophysical head model.

2. For each neural population in the brain network model, we calculate the perturbation caused by the electric field as explained in our post about Predicting how brain stimulation affects neurons. 

3. We simulate these perturbations in the personalized brain network model.

4. We quantify the spread by counting the number of brain regions to which the seizure propagates.

   
   

This evaluation relies on the imprint hypothesis, which is the idea that the acute effects observed in the model (e.g., suppressed spread) can predict long-term therapeutic outcomes.

The optimal montage is the one that reduces seizure spread the most. But here’s the challenge: many different montages might suppress seizures equally well. So, how do we choose among them?

To resolve this ambiguity, we incorporate a secondary objective inspired by our tES 1.0 approach: a target map. This map prioritizes inhibitory stimulation in both EZ and PZ regions. Among all montages that stop the seizure, we prefer those that also best match this target map.

This multi-objective optimization is not trivial. To navigate the complex, high-dimensional montage space, we use an evolutionary algorithm—a method well-suited for searching under uncertainty and non-convexity.

Looking ahead

This Neurotwin-driven pipeline is already being used in an ongoing clinical trial (NCT06334952). As we continue to refine these tools and collect clinical evidence, we hope that this methodology will enhance the safety and effectiveness of tES. We believe that with the Neurotwin framework, we can:

• Identify new targets beyond the obvious epileptogenic zones.

• Enhance the safety of the therapy by avoiding unintended excitation of sensitive areas.

• Provide a rational, model-based framework to better understand the network-level effects of tDCS.