We are about to start using fmriprep to process tens of sessions of functional data collected for a given subject, all in the same scanner (Prisma) and with the same sequence. There is a separate, older multi-session dataset collected on the same subject in a different scanner (Skyra), with a similar sequence and spatial resolution. Is it advisable to combine the two datasets into a single fmriprep run, or would it be better to process them separately and later register the second to the first?
FMRIPREP does not perform any operations across participants (in other words it does not have a “group” BIDS Apps analysis level) so it does not matter if you keep participants from two sites in one dataset or two separate ones.
In context of BIDS you can read about organizing multisite data in section 10 of http://bids.neuroimaging.io/bids_spec1.0.2.pdf.
Thank you for the answer and the pointer! The situation here is a bit different - it’s always the same participant, but multi-session data in each of the two sites. What I was wondering is whether the handling of multi-session would be problematic if combining data from the two sites.
In such case I would recommend using the --longitudinal option http://fmriprep.readthedocs.io/en/latest/usage.html#Workflow%20configuration. Please mind that in this dataset no matter what processing you do site effects will be very hard to distinguish from session effects
Thank you! I think that, in that case, I won’t try to process both sites together (it’s already longitudinal within sites).
Well, if you want all sessions to be in the same native/T1w space processing them together makes sense. The --longitudinal flag will make sure that the participant specific template (estimated by combining all T1w’s of the participant) will not be biased towards one session. This template is being used to realign all functional runs (across sessions).
If you plan to use MNi template space it might be worth using the --output-grid-reference to make sure all outputs will have the same field of view.