Hello all! I have a question regarding group maps to share in a BIDS derivative dataset. I couldn’t find anything about group average statistical maps being stored in BIDS derivatives so I am asking this question out in the open here…
Is there a specification or guidance anywhere on derived BIDS imaging maps? I want to share a group averaged NIfTI of a metric (rates of cerebral protein synthesis), but where would I put this in the derivatives folder structure? For some of the studies I’m preparing I have upwards of four different “conditions”. Since it’s a group map, it is a composition of many participants’ data and thus would not receive a sub- entity/label. Any thoughts or ideas are welcome. Thanks!
From what I can tell, BIDS Stats Models only describes the models that generate model outputs, with no recommendations for the actual outputs. However, maybe fitlins has some useful conventions in place? I started working on formalizing fitlins’ conventions in a (currently unnumbered) BEP, but that BEP is very preliminary.
@tsalo Thanks! I love it! That’s what I think I’m going to do for now, is follow that unnumbered BEP and put the group maps in a group/ subfolder under the /derivatives/pipeline_name/. I’ll leave this open for now in case anyone else has comments, but if no other comments come through for a while, I’ll try to remember to accept your solution.
@tsalo
Am I correct that you also tried to align the output of nilearn’s GLM with something more BIDSish as well?
Also something that would be good to keep in mind is leveraging the nidm results format as much as possible for sharing statistical maps: it is a standard that already exists, interoperable with several of the main fMRI softwares and is neurovault friendly
That’s right. I tried to ensure the nilearn outputs were similar to fitlins’ outputs, using that unnumbered BEP as the synchronization point.
The Nilearn devs did discuss adding NIDM-Results support to Nilearn a year or so ago, but we decided that, given how detailed (and complicated) NIDM-Results is, the amount of extra work that would require was too much. Instead, we went with the BIDSish naming convention and json sidecar files as a compromise. If there’s an external library that could be used in conjunction with Nilearn to write out NIDM-Results, I’d happily use it, but the last time I checked, nidmresults wasn’t under active development.
Thanks @tsalo and @Remi-Gau for the great context and some history! I will be able to move forward with this now and have added some comments to the unnumbered BEP too. If I’m feeling really ambitious, I’ll add more to that unnumbered BEP using fitlins and NIDM-Results as guidance.