I would like to run the pre-trained mriqc classifier on a large dataset that I have preprocessed with fmriprep. From the mriqc workflow it appears that most of the steps done by mriqc to generate the feature vector have already been done by fmriprep. Is there an easy way / a module that gets me from a fmriprep output folder to a group level tsv file with the input features for the mriqc classifier? My dataset is rather large so I’d like to avoid re-running steps if at all possible.
For the BOLD images we don’t have a RF classifier yet, so I guess you are referring to the anatomical images.
In that case, you would only missing the “hat” mask, which you could easily bring into the T1w image via the MNI template transform.
Very recently we have started to detach the structural MRI processing from fMRIPrep. We would really appreciate your contributions on sMRIPrep - https://github.com/poldracklab/smriprep
Sadly, I can’t say there is something already implemented :(.
Thanks so much for all of the work you guys have done on developing fmriprep.
We’ve just run our entire resting and task fmri dataset and we’re in the process of deciding which artefact/motion regressors to include for first level analysis. Is it possible to run MRIQC after fmriprep on functional data? It looks like from the previous post that there may be a ‘combined’ pipeline in process for fMRI, but it isn’t ready yet?? Any advice on how we could implement MRIQC with our fMRI data processed through fmriprep would be much appreciated. E.g., could we run MRIQC separately and then take the output and use it for our first level analyses?
MRIQC is for raw input data (i.e., before processing)
For the quality assessment of fMRIPrep results, please check the visual reports it generates (tutorial)
Thanks for this clarification! I am looking at the group reports generated by MRIQC; is anything like that available to summarize fmriprep QA reports over a group of people/sessions? @effigies
Right now, we don’t have anything like that, unfortunately. This is something we’re looking into designing with @Shotgunosine and the ABCD folks, as there’s no feasible way to review 10K subjects manually, so hopefully we’ll have something soonish, but we don’t have a timeline yet.