Fixing bad brain masks and T1s when using fmriprep-24.1.1

Software Support
1

Hello,
I’m running into a couple of issues with fmriprep and I wanted to get some clarification on the pipeline.
One of the issues that has been coming up is that for some participants, the brain mask either excludes huge chunks of cortex or it includes void space outside the brain. Sometimes the brain mask will fix itself if we re-run fmriprep. Sometimes, it comes out with other issues.

Another issue is that for a few of our participants, the average T1 image created by fmriprep (an average of two T1w images) looks blurry, which affects the contrast and subsequently the tissue segmentation. This issue seems to happen randomly–two researchers in our team ran the same subject in separate instances and one got a good T1 and the other got a blurry T1. We also tried running the stand alone freesurfer function mri_robust_template and the resulting averaged T1 was not blurry.

So far, we have been running fmriprep 24.1.1 with the anat-only flag to run some quality control checks before running the functional preprocessing. My question is, if we get errors with the brain mask, can we fix the brain mask outside of fmriprep and run the full fmriprep pipeline (anat + func)? Will the manually edited brain mask be used or will it be remade? Will manually editing the brain mask cause issues downstream?

Similarly, if we get a blurry T1 image, can we create one with the mri_robust_template function outside of fmriprep and replace the blurry T1? If we do this, will fmriprep re-do the tissue segmentation using the good T1 image?

Thanks for any information you can provide.

2

Hi @dianaperez25,

As of recent fmriprep versions you are able to pass in precomputed derivatives, however they are made, as long as they follow expected BIDS naming conventions.

Best,
Steven

3

Thank you, Steven. Is there anything that needs to be specified to fmriprep? I ask because we tried to replace the bad brain mask with a fixed brain mask using BIDS naming conventions, but fmriprep still re-made the brain mask. So we were wondering if there was anything additional we needed to do in order to pass those precomputed derivatives.

Best,
Diana

4

Hi @dianaperez25,

Please refer to the documentation, noting that this behavior was improved in 25.0.0: What’s new — fmriprep version documentation

Best,
Steven

5

Thank you. Unfortunately, it is not possible for us to upgrade to the newest version of fmriprep at this time. Is there anything to be done to pass precomputed derivatives to fmriprep 24.1.1?

Best,
Diana

6

@dianaperez25,

The --derivatives flag should still be available to you after 23.2.0. You can look at previous versions of documentation https://fmriprep.org/en/24.1.1/usage.html#reusing-precomputed-derivatives

Best,
Steven

7

Reusing anatomical masks should work just fine in 23.2+, as should most anatomical derivatives (with an exception of tissue probability maps: Handling of white matter lesions in fMRIPrep - #2 by pruittpj). Reusing preprocessed fieldmaps and functional derivatives should work correctly in 25.0+, though bits and pieces might work well enough in older versions.