Duneuro forward modeling using containers

This tutorial explains how to compute EEG/MEG forward models with duneuro

using the new container-based integration in Brainstorm.

Using containers allows running duneuro without manually installing

its dependencies or compiling external libraries.

---

Introduction

Duneuro is an advanced forward modeling tool for EEG and MEG.

It relies on a number of external numerical libraries, which can make

manual installation difficult.

Brainstorm now supports running duneuro using containers,

which package duneuro and all its dependencies in a reproducible environment.

This tutorial demonstrates how to use duneuro through containers

directly from Brainstorm.

---

Requirements

• Brainstorm (development version including container support)
• A supported container runtime:
  • Docker (Windows, macOS, Linux)
  • Podman (Linux)
  • Apptainer / Singularity (HPC clusters)

If you have not installed a container runtime yet, please follow:

?Using Containers with Brainstorm

---

About containers (short reminder)

Containers allow Brainstorm to run external tools without manual installation

of libraries or compilers.

Note: Docker requires a running daemon (a background service).

Podman and Apptainer do not.

---

Dataset

This tutorial uses the same example dataset as the standard duneuro tutorial.

You may either:

• Follow the tutorial using your own subject anatomy, or
• Reuse the sample anatomy provided in Brainstorm tutorials

The exact dataset is not critical for demonstrating the container workflow.

---

Step 1: Prepare the anatomy

As with any forward modeling approach, you need a subject anatomy.

1. Import a subject MRI
2. Compute the cortical surface
3. Import or generate the head surfaces (scalp, skull, brain)

Refer to the general Brainstorm documentation if needed:

?Anatomy processing

---

Step 2: Open the forward modeling interface

1. Select your subject
2. Right-click on the subject name
3. Select:

   • Compute head model

In the head model computation panel:

• Select duneuro as the forward model

• Select EEG, MEG, or both, depending on your data

---

Step 3: Enable container-based execution

In the duneuro options panel:

• Enable the option to run duneuro using containers

• Brainstorm will automatically:
  • Detect an available container runtime
  • Check that the runtime is usable
  • Download the duneuro container image if needed (after confirmation)

No manual container commands are required.

---

Step 4: Container image download

If the duneuro container image is not already available, Brainstorm will

display a confirmation dialog asking whether the image should be downloaded.

After confirmation:

• The image is downloaded automatically
• The image is cached and reused for future runs

This step is required only once.

---

Step 5: Compute the forward model

After the container is ready:

1. Validate the duneuro options
2. Start the computation

Brainstorm will run duneuro inside the container and retrieve the results

automatically.

From the user perspective, the workflow is identical to a native execution.

---

Inspecting the results

Once the computation is complete:

• The head model appears in the database
• Sensitivity maps and lead fields can be visualized as usual

For MEG:

• Tangential sources show higher sensitivity than radial sources
• This behavior is expected and reflects the physical properties of MEG

---

Advantages of the container approach

Using duneuro with containers provides several benefits:

• No manual installation of duneuro dependencies
• Identical behavior across operating systems
• Easier reproducibility of results
• Simplified usage on shared systems and clusters

---

Notes and limitations

• Container support is currently considered experimental

• The interface may evolve based on user feedback
• Future versions may integrate container management more tightly
  • with the Brainstorm Plugin Manager

---

Summary

• Brainstorm can run duneuro using containers
• Only a container runtime is required
• Duneuro dependencies are handled automatically
• The workflow is identical to standard head model computation

---

See also

• Standard duneuro tutorial

• ?Using Containers with Brainstorm

• Container support PR