I am trying to use MNE to do the source localization for EEG recorded from patients who have done cingulotomy. Basely parts of their brain were removed.
Could anyone tell me if they have done this before? Will the results be wrong? Of course, the MRI was taken recently.
Using freesurfer, I have recon -all the head models. From the view, I could observe that parts of the brain were missing.
Thank you.
If you have a question or issue with MNE-Python, please include the following info:
MNE version: e.g. 0.24.0
operating system: e.g. macOS 12 / Windows 10 / Ubuntu 18.04
To get the correct formatting, paste your Python code, then select it, and click on the Preformatted text toolbar button.
Please avoid sharing screenshots of code and error messages! Screenshots are difficult to work with especially on mobile devices; they don’t allow others to copy code and other text; and they don’t show up in the search.
Please edit or remove the above text before submitting your posting.
@yourmaverick - it would be difficult to do this correctly, but it depends on how the surgery was done. In MEG, epilepsy patients with prior resections are routinely analyzed. In EEG, however, the skull plays a much much larger role in the forward model (40X the resistance of other tissues). Typically, you will have a breach artifact that can be seen on the channel recordings because the electrodes near the craniotomy see a larger signal (because of lower local skull resistance).
There was some previous research from John Ebersole out of University of Chicago with ECOG grid and scalp EEG (for scalp dipole modeling confirmation vs grid data). I think they used bone wax to seal the craniotomy and prevent breach artifact during the invasive monitoring. But after the surgery is performed, you are still likely to have a defect in the skull where the craniotomy was performed.
The accuracy really boils down to how much breach artifact you are dealing with. There are laser based surgery methods (particularly for midline surgeries) that create a smaller defect in the skull, so maybe this would produce a more accurate inverse along the lines of a non-surgical subject.
If you ignore the channels with breach artifact (I don’t know how you would determine this), you could do the modelling. Or you could model the broken skull portion as a separate compartment (may need to use Finite Element Modelling) - but it would be hard to know what resistance value to give as different patients may heal differently.
A lot of it depends on the surgery and the signal that you are trying to find. The further away from the surgical site the better. Hope that helps.
Thank you so much for your reply.
From my understanding, your concern is the artifacts in the EEG data, not the inverse problem.
If I could clean the EEG data, I might be able to do the source localization., right?
Let’s assume the EEG data is clean. If part of the brain tissue is removed, how could the EEG be projected to that specific location?
So one thing that seems ambiguous - was the EEG taken before or after the surgery? My previous comments were based on an EEG taken after surgery.
But based on:
It sounds like the EEG was done before the surgery, otherwise the activity should be theoretically 0 at the removed area. If this is presurgical EEG - just treat it as normal for the preprocessing/forward/inverse.
Scenario 1 - no/minimal post surgical brain shift. Try an affine coregistration of pre/post MRI (fsl/spm/freesurfer/afni/dipy all do this easily). After coregistering the two, create the freesurfer surfaces and BEM based on the pre-surgical MRI and use this as your source for EEG modelling. Once you have the source activity, you will need to create your region of interest - which is the surgical region. This can be generated in mricron or 3Dslicer (probably others too). Once you have the surgical mask, you can just multiply this by the activity image (masking out the activity in the surg region). Using surfaces may make this more difficult to do compared to volumetric.
Scenario 2 - significant post surgical brain shift. This is much more difficult. You will need to use non-linear warping. The problem is that you are coregistering two objects with different amounts of brain tissue. So the warping is inherently incorrect or at least very difficult. I don’t know of any easy way to do this. ANTS - may have some functionality to do this. It is supposed to be able to coregister atypical brain volumes.
There may be others that have more knowledge on this. @alexrockhill has done a lot of work with dipy and might have some more info.
Thank you so much for your detailed explanation. To be honest, these are pretty new to me. It would take a lot of time to understand them.
Our EEG data was recorded after the surgery. It looked pretty good (not close to zero). I am wondering if I still need to combine MRI taken before and after surgery to build the model ?
From my experience, even though there are different amounts of brain tissue, MRIs or MRI-CT registrations tend to work pretty well. I think this is because there is a lot of information about the alignment based on where the skull is and that is what really drives the optimization.
Generally, you find the electrode contact positions on the CT, register it to the pre-surgical MRI and label the electrode contact positions based on that pre-surgical MRI. You could register the pre-surgical MRI to the post-surgical MRI and label contacts based on that. If the brain shifts back at all (since the CT with the positions is intraoperative), the gains might be marginal, however. I think in general, you could get pretty advanced in modeling the brain shift but, since the contacts are on the order of a few millimeters in diameter, you might be within the margin of error of the recording anyway.
If you have a question or issue with MNE-Python, please include the following info:
Please also provide relevant code snippets – ideally a minimal working example (MWE).
To get the correct formatting, paste your Python code, then select it, and click on the Preformatted text toolbar button.
Please avoid sharing screenshots of code and error messages! Screenshots are difficult to work with especially on mobile devices; they don’t allow others to copy code and other text; and they don’t show up in the search.
