split fmriprep anat from func?

[faskowit]

With the LTS release finalize soon... might be an opportunity to also consider a slight change in how fmriprep brainlife is run. There is an option to run only the anatomical part of fMRIPrep: https://fmriprep.org/en/stable/usage.html#reusing-precomputed-derivatives , and then to feed in this anatomical output to the functional processing. I could imagine this would massively cut processing times for datasets with more than one functional scan. What do you think? It would require a sort of raw frmiprep-anat datatype... that would be unzipped when the functional fmriprep is run.... does that make sense?

[soichih]

I think it makes sense to create a new datatype for anat-derivatives. I will create the datatype (experimentally) and update the fmriprep to store output to it.

I am guessing that freesurfer is really the big chunk of the anat preprocessing. If so, then what we have right now (having freesurfer as a separate step) would probably be sufficient to reduce repeated anatomical preprocessing so we might not need to use this functionality?

[giulia-berto]

What if we allow the user to insert multiple inputs for the func/task? I don't know if it makes sense, or if the output will be to complex to handle.

[soichih]

@giulia-berto Yes, BL currently can't handle varying number of output objects.. so that won't work. Even if we could.... user could be feeding 5 bold images, or 500 bold images, so it becomes difficult to predict how much resource we need to run fmriprep. By processing 1 bold at a time, BL can distribute the processing across multiple jobs / clusters.