As you can see, and related to a question I posted recently (Parcellation for MNI152NLin2009cAsym? - #14 by PhoenixByrne), my T2w outputs are in native space. In response to this post, it was suggested that I resample the atlas and the standard to my native space resolution in order to create masks. I do have a few questions however:
It appears preferable to resample the atlas and standard above resampling the functional image, is there a reason for that? does resampling the functional image after pre-processing cause problems down the line?
In my conversation in the previous post it was highlighted that the HOC/S atlases don’t have a .tsv to aid in identifying brain regions. Even if I used other .tsvs provided by the templateflow e.g. * tpl-MNI152NLin6Asym_atlas-HOCPA_dseg.tsv to get a broad idea, won’t this become increasingly redundant to use once I resample the atlas and standard?
Finally, if I manage to generate masks from resampling the standard and atlas and apply them at the single subject level, won’t this cause problem in group-level analysis when it looks to register everything to a standard space?
Apologies for cross-posting, I wanted to avoid my question getting lost in the previous thread. Any clarity would be greatly appreciated, I can’t seem to get a consensus in my department. I do want to note that it’s currently not possible for me to re-run this preprocessing, I appreciate that most of my problems could be solved that way!
Computational efficiency: it is easier to perform operations on a binary mask than a full timeseries image.
Minimizing warps: fMRIPrep is nice in that all interpolations and registrations are performed in a single step, minimizing how many times data need to be interpolated. Performing registrations after fMRIPrep means additional data interpolation, which is not preferred. Since atlases are just a collection of integer-valued masks, there’s really no interpolation to worry about.
Not sure what you mean by redundant here. If anything, those will still be helpful because the index-to-region correspondance shouldn’t change between spaces ( e.g., index 7 will always be temporal pole).
Depends. If you are doing a classical whole-brain group analyses then yes you need everything in a common space. As you stated, all of these issues could be resolved by preprocessing again and specifying MNI152NLin6Asym as an --output-space (I also would not analyze data processed with the --fs-no-reconall, for future reference). But if you are abstracting a different kind of data, e.g. atlas-based connectivity, then you do not necessarily have to have all brains aligned.