I used the T1w anatomical mask from fmriprep as an explicit mask for a GLM - should I have used the func mask?

Good morning/afternoon/evening to everybody in your various time zones!

My question is as stated in the title :slight_smile: I know the BOLD and the T1w images are in the same space, but the masks clearly cover different regions of the brain, at least on the extremities (see the image posted below). The top mask is the T1w from the ‘anat’ folder, while the bottom is the BOLD mask from ‘func’ - the func mask clearly extends past the anat mask.

Without thinking I generated some first level models using the T1w mask as an explicit mask, but in hindsight, it seems as though the functional BOLD mask might have been more appropriate. I want to make the right choice!

I’m at this very moment attempting to comb through some papers and see what others have done, but mask parameters do not seem to be commonly reported, and doing an outright internet search isn’t returning much either. Any thoughts are appreciated!!!

As always, I appreciate the support I’ve gotten from this community, and I value the time you take to respond (I know it’s finite).

Kade

I can think of two points of view:

  • if you want to analyze that particular individual and only that one, I’d take the most liberal mask: it costs you basically nothing to be a bit more generous.
  • but most often you deal with different subjects, and in that case, I advise not to use subject-specific masks: this makes inter-subject comparison/analysis much harder. Using a standard MNI-space brain mask is reasonable, or some statistical consensus among the individually defined masks.
    HTH
1 Like

I think that is normal. Functional mask may not be well regenerated. We usually refined it with anatomical mask which is always very good.

1 Like

Thank you for your response! I appreciate getting multiple points of view on how to address the issue.

We’ll be conducting a searchlight analysis on individuals, then averaging those accuracy maps at the group level for statistical testing. It sounds like, since we plan to go to the group level, a standard MNI mask would be more appropriate?

When you say “statistical consensus” do you mean something like an average anatomical mask?

It sounds like, since we plan to go to the group level, a standard MNI mask would be more appropriate?
-> Yes

When you say “statistical consensus” do you mean something like an average anatomical mask?
-> Yes. Nilear has a routine for that.
http://nilearn.github.io/modules/generated/nilearn.masking.compute_multi_epi_mask.html
http://nilearn.github.io/modules/generated/nilearn.masking.intersect_masks.html

Best,
Bertrand

1 Like

Thank you again, Bertrand!

I think I will likely use a mask from the standard MNI template everybody was warped to, as it seems to map to the individual subject data pretty accurately when I overlay them. It is good to know about those Nilearn routines, though, so I appreciate the advice on that!