Source analysis for full trial vs segmented trial

Discussions
,
1

Hi All,

I have data from an EEG experiment that includes a variety of simple limb movements (e.g., hands, ankle, etc.). The experiment was a block design repeated multiple times. The experiment was also repeated in the fMRI scanner. I am trying to use ROI results from fMRI analysis to create custom scouts. I have digitized EEG electrode locations and subject anatomy. These have already been imported into the toolbox. I have two questions related to this topic:

  1. Is it better to run source analysis for the full trial (~8 min total) and then segment the results based on movement (12 second movement windows), or segment each movement window (e.g., ankle window, hand window, etc.) and solve the inverse problem for each one? Or, does it even make a difference at all?

  2. Are you able to point me to a good resource on how to export ROIs from freesurfer into brainstorm? I saw the tutorial on scouts but I wasn't clear on what format Brainstorm is expecting.

Please let me know if any aspects of my question were unclear.

Thanks,

Justin

2

Is it better to run source analysis for the full trial (~8 min total) and then segment the results based on movement (12 second movement windows), or segment each movement window (e.g., ankle window, hand window, etc.) and solve the inverse problem for each one? Or, does it even make a difference at all?

It does not make any difference, the minimum norm solution is independent from the recordings. But you can obtain different results depending on the data you used to estimate your noise covariance matrix.
https://neuroimage.usc.edu/brainstorm/Tutorials/SourceEstimation#Ill-posed_problem

Are you able to point me to a good resource on how to export ROIs from freesurfer into brainstorm? I saw the tutorial on scouts but I wasn't clear on what format Brainstorm is expecting.

Using the volume source models, you can import your ROIs defined from the fMRI as .nii or .mgz files:
https://neuroimage.usc.edu/brainstorm/Tutorials/TutVolSource#Volume_atlases

There are probably solutions to project your fMRI results on the cortical surface with FreeSurfer, but I think there is very little chance that this ends up giving you the regions where the minimum norm estimation locates the corresponding activity.

Note that fMRI and EEG record completely different mechanisms that might not have a direct correspondence, and with completely different temporal dynamics. You might not observe the same things with the two modalities, therefore constraining one with the other is not always meaningful. If you want to perform this kind of analysis based on the fMRI, I'd recommend you also perform another test where you define the ROIs from the data (=where you observe the peak of activity you are interested in).

One very interesting way of combining fMRI and MEG is illustrated in this article:
https://sci-hub.tw/https://www.nature.com/articles/nn.3635

3

Hi Francois,

Thank you for your detailed response. You definitely answered my questions. As for constraining the analysis, I was not planning to constrain the inverse solution, but rather to look at the scout based on the fMRI ROI. Perhaps this is still what you were saying. Do you mind to clarify what you mean by this:

I'd recommend you also perform another test where you define the ROIs from the data (=where you observe the peak of activity you are interested in).

Do you mean peak of activity in the fMRI or EEG? I definitely understand that its possible that the fMRI and EEG activations may not give corresponding results. I have seen that paper before and will look to see if their approach can be implemented in my analysis.

4

Do you mean peak of activity in the fMRI or EEG?

It is a good idea to define scouts from the fMRI results.
However, because you might observe different things in the fMRI results and in the source-reconstructed EEG, it might worth it looking at where the peaks of the EEG are. Make sure that there is a spatial overlap between where you observe the "possibly meaningful peaks" of the EEG source maps and your fMRI blobs before you run all your analysis.
Another way to formulate this recommendation is: look at your data before processing it fully automatically.

5

Ah, I see what you are saying. Quite simple but very important! Yes, I have explored the data (e.g., topoplots, averaged data, etc.) I will definitely compare this to the fMRI results in more detail.

Thanks again!