I have multi-echo fMRI data that I’ve converted using dcm2bids and want to do a limited pre-processing with fmriprep.
Is it possible (desirable?) to turn off the ANTs registration to the MNI template within fmriprep, and have it “just” perform the following: motion correction, slice timing correction and generation of the optimally combined (OC) timeseries/t2star map?
I’d then use that as input to tedana to denoise the OC timeseries, and input the result to FSL, using their registration pipeline to the T1w and MNI standard. FSL’s approach was always to process fMRI data in acquired (motion corrected/undistorted) space, and I’ve always used that - doing the registration to standard space when needed later for group analysis.
Or should I just apply the transforms to get what I want at the end? Just wary of compounding interpolation errors.
fMRIPrep only writes out the OC time series to MNI space. If you want individual echoes to run through tedana, then you can use the --me-output-echos option. That will output minimally-processed (STC+HMC+SDC) echo-wise files that you can feed into tedana.
From there you can either denoise the OC data in MNI space or use the denoised subject-space data from tedana and apply the transforms to MNI space to that file. It sounds like you’d prefer to do the latter, which is entirely reasonable. I personally prefer the former, since the number of transforms applied to the data is the same and it’s just easier to denoise data in the target space than to apply transforms after denoising, but either way is fine.
There’s no reason you can’t use the transforms from fMRIPrep to register your denoised data to standard space. I don’t think there’s any need to calculate different registration transforms with FSL unless you have a particular approach that works better than fMRIPrep on your data.
The standard application of tedana involves (1) estimating S0/T2star, (2) calculating the OC time series, (3) applying ME-ICA to generate ICA components that can be used to denoise the data, and (4) denoising the OC data with those ICA components. Tedana will output both the OC data without denoising (desc-optcom_bold.nii.gz) and the denoised OC data (desc-denoised_bold.nii.gz). You can also use the ICA components generated by tedana to denoise your OC data in a different space (e.g., the MNI-space OC data produced by fMRIPrep), which is what I typically do, or you can use the denoised data and register it to another space.
Basically, the answer to your first question is “yes”, and the answer to your second question is “you can denoise the OC data or let tedana do it, and OC and denoising are not mutually exclusive.”