I understand that turning freesurfer recon off in fmriprep may lead to less accurate registration of EPI to T1 structural data as boundary based registration (BBR) is not used. Has anyone compared the results from this “fmriprep:BBR” to BBR as implemented in FSL? I guess one significant difference would be the approach taken to find the grey/white matter boundary - FAST vs freesurfer.
As a related question, if I do use freesurfer recon, and have fmriprep do all the heavy lifting for EPI → T1 structural → standard brain registration, is there a way to take the generated transforms/warps and build a “reg” folder that would keep FSL happy?
For various reasons I have to keep the functional data in “func” space for the purpose of processing (tedana, 1st level FEAT), but would then like to take advantage of fmriprep’s registration pipeline when performing group analysis.
Sorry for the slightly vague questions, if someone can point me in the right direction I’d be most grateful.
Not explicitly, as far as I know. But empirically surfaces from FreeSurfer tend to look better than those from any volumetric workflow. For example, the difference between MRtrix’s 5ttgen (5ttgen — MRtrix3 3.0 documentation) for the fsl (no surface data) and hsvs (uses freesurfer) options is pretty significant.
You can have both a func and MNI output space in fmriprep. Then all necessary xfms will be saved out and you can concatenate them as needed. But doesn’t tedana and FEAT work with MNI images too?
Thanks for getting back to me - good to know that there are likely to be some differences in quality of the BBR registration.
As for tedana/FEAT, I want to have the input data minimally processed before running tedana, just motion and slice-time corrected. Once the data have been “cleaned-up”, I can then use the xfms to normalise the data to template space - and input that to FEAT (as you suggest).
Q. Are there any issues with smoothing functional data already transformed to MNI output space? I know FSL does smoothing early in the pipeline (in native/func space), not sure about SPM.
Last question, I’ve previously run fmriprep as described above (with --fs-no-reconall, with --output-spaces func). If I now wanted to generate the necessary xfms (using freesurfer) that would allow me to then transform already cleaned-up tedana output to the MNI output space, can I have fmriprep do that? I’d rather not have to go through the manual classification for tedana again…
I believe that smoothing is typically done after spatial normalization (if normalization is part of the pipeline). I believe this is because non-linear warping introduces smoothness into data. So if you smooth first and then normalize, you add additional smoothing that can vary across people. If you normalize first, you can dictate the final smoothing amount.