Dimension mismatch between fMRIPrep and templateflow for MNI152NLin2009cAsym

, ,
1

I have preprocessed my data using fmriprep 24.1.1. with default parameters for output space.

The T1w output dimensions match with Templateflow’s tpl-MNI152NLin2009cAsym_res-01_desc-brain_T1w.nii (at 193x229x193; 1x1x1)

However, The bold (and boldref) output dims are 77x95x82; 2x2x2 compared to Templateflow’s tpl-MNI152NLin2009cAsym_res-02_desc-fMRIPrep_boldref.nii (and T1w and T2w) at 97x115x97; 2x2x2 .

bold output filenames end in _space-MNI152NLin2009cAsym_desc-preproc_bold so it seems that fmriprep thinks they are in that space. There hasn’t been any resampling in terms of resolution either, as my EPI scan had 2x2x2 voxel size.

What is going on?

2

Hi @Mkassaie,

I believe this is because fmriprep is maintaining the field of view of your native image. Specifying the res modifier may change this behavior.

Best,
Steven

3

Thanks @Steven ! I also suspected that to be the case.
Oddly, when I gave a space (e.g. MN152NLin6Asym) without determining a res, the bold dimensions matched tpflow’s MN152NLin6Asym dimensions.

My native image is 104x104x72; 2x2x2
fmriprep default output is 77x 95x82; 2x2x2 (not matching MNI152NLin2009cAsym)
fmriprep MNI152_6th output is 91x109x91; 2x2x2 (matching MN152NLin6Asym)

So it seems that fmriprep upsamples my image to fit the template space only if it is something other than the default MNI152NLin2009cAsym.

4

Hi @Mkassaie,

Interesting. I am guessing that this may be because the default MNI2009c transform is used for other steps in the process, so it may always be calculated. The resampling to voxel resolution / FOV matching may only happen if an additional transformation is requested?

Best,
Steven

5

@Steven I wanted to give an update. I think my issue is just the bounding box. I just need to resize(i.e. add zero-padding) to match the MNI dimensions. (This thread seems to be about the same issue).
I’m not sure how to do this without warping or resampling my image. I tried different settings with MCFLIRT but I had no luck. I guess something like mrgrid that you recommended here may be what’s needed, but I don’t wanna install a whole toolkit to this. This also sounds like a neat trick that would accomplish something similar, but i don’t know whether the image will end up being centered the same way as the MNI template.