SEEG Reference, Montage, and Source Modeling: Questions and Reflections from the Toulouse Workshop[MicMac2025] and Beyond

Discussions
, , ,
1

Dear all,

This post collects and clarifies some technical discussions and questions raised during and after the Micmac 2025 workshop in Toulouse, with a focus on SEEG processing in general and using Brainstorm in particular.

The topics covered—reference handling, montage choices, and source modeling—are foundational to accurate interpretation and reproducibility in SEEG research.

The answers reflect both individual expertise and collective input from the Brainstorm team, including insights from Dr. Mosher, Yash Vakilna, and others. Wherever possible, responses are supported by forum discussions, code references, and peer-reviewed publications.

We hope this compilation serves not only to document best practices but also to foster open dialogue and community-driven refinement. Feedback, corrections, and additions are welcome.

1. SEEG Reference Handling

Q: How is the reference in SEEG data handled in Brainstorm?

  • Answer: In SEEG recordings, the reference electrode is often defined by the neurosurgeon and may not be recorded explicitly. In practice, once the data is imported and re-referenced (e.g., to an average or bipolar montage), the original reference becomes irrelevant. For forward modeling, a single reference electrode is often selected. For inverse solutions, Brainstorm typically uses an average reference, and the gain matrix is re-referenced accordingly. This is confirmed by the Brainstorm forum and codes.
  • Please refer to this detailed explanation by Dr. Mosher on the Brainstorm forum: https://neuroimage.usc.edu/forums/t/reference-of-the-gain-matrix-eeg-openmeeg-bem-in-brainstorm/1248/7

2. SEEG Montage and Source Localization

Q: Why use the “origin” montage instead of bipolar montages for source localization?

  • Answer: No matter which montage (like bipolar or average reference) you use, as long as you’re using the same set of electrodes, the source modeling steps treat those channels in a way that effectively “undoes” any simple scaling or combining you did beforehand. In other words, rearranging the channels or weighting them differently (a “linear” change) doesn’t affect the final picture of where signals are coming from in the brain. The only thing that really matters is which channels (i.e., which electrodes) are included or excluded.
  • See the above forum note from 2014 for the mathematics. The result is that source estimation is agnostic to the linear weights used in any sensor montage, with the key consideration being simply which sensor region you have selected (“good” vs “bad” channels).

3. Fingerprint Analysis and Montage

Q: Have we tried different montages (besides bipolar 2) for fingerprinting?

  • Answer: “At the sensor level, we’ve experimented in-house with referential fingerprinting, but in our Brain publication, we focused on bipolar source montaging. This bipolar approach highlights local sources by examining signals from pairs of adjacent electrodes.
  • At this workshop, we demonstrated a ‘source’ modeling approach, which goes beyond any sensor montage variations and directly uses a dipolar source model. We then ran the same time-frequency fingerprint analysis on this dipolar representation and confirmed that the same underlying pattern remains.
  • As we discussed above, source modeling is agnostic to any particular sensor montage.
  • Original publications:

4. Forward Model BEM Surfaces

Q: Is it necessary to include the scalp and skull in the BEM for SEEG data?

  • Answer: No, it's most likely not necessary. Since SEEG electrodes are intracranial, only the inner skull surface (brain boundary) has a primary effect. Including the scalp and skull adds probably unnecessary complexity. This is supported by Dr. Mosher and foundational modeling work like Hämäläinen & Sarvas (1989) [Realistic conductivity geometry model of the human head for interpretation of neuromagnetic data - PubMed], which shows that extracranial eeg can be approximated with a variation of an “isolated skull.” Thus intracranial data can also be simplified.

5. Source Model Choice (Surface vs. Volume)

Q: Why use a cortical surface model instead of a volume grid for SEEG source imaging?

  • Answer: The choice is driven by research goals. Cortical surfaces are preferred for:

    • Interpretation aligned with the assumption that sources originate in gray matter.
    • Compatibility with minimum norm estimates (MNE), which are classically surface-based.
    • Better visualization feature within Brainstorm.

  • Volume grids can be used, especially with beamformers, but may require projecting deep white matter activity back to the cortical surface, adding complexity for interpretation.

Q: Why MinNorm/sLORETA over other inverse methods?

  • Answer: The choice was practical and based on interest in time-series analysis. While methods like dSPM or sLORETA apply different statistical weights, they mainly differ by scale factors in the resulting time series. For time-frequency (TF) analysis, this difference may be negligible. The selection remains a research-driven decision.

6. SEEG Contact Localization from Curry

Q: What coordinate system does Curry use, and how is it handled in Brainstorm?

  • Answer: Curry uses its own coordinate system.
    When importing Curry channel files into Brainstorm:

    • Brainstorm supports Curry’s format natively (.res;.rs3;*.pom). Users can read Curry output files.
    • For Curry, or if the original coordinate system is known, users can use Brainstorm functions like 'cs_convert' to map the coordinates to the Brainstorm native coordinate system (SCS).

  • If the conversion is not known, coregistration can be achieved by aligning scalp points (digitized head shape) with the MRI-derived surface.

  • In all cases, manual adjustment and visual inspection are recommended to ensure accurate alignment using Brainstorm’s built-in visualization and editing tools.

7. sEEG Data Availability

Q: Can more sEEG-marked datasets be shared?

  • Answer: At this time, only exemplar datasets are publicly available.

Thanks to Dr. Mosher, Yash Vakilna, and the BST team for reviewing and collecting responses to these questions.

2 Likes