Want to use ACompCor in Fitlins, but no Cosine Regressors from fMRIPrep

Summary of what happened:

Hi all,

I preprocessed task fMRI data with fmriprep 22.1.1 with no errors. However, there were no cosine regressors output with confounds, perhaps because it was a relatively short run (154 volumes at TR=800ms, whereas longer runs did have cosine regressors).

I would like to run GLMs through fitlins (0.11.0) using an acompcor-based model, but I do not have the appropriate cosine regressors to high-pass filter correctly. I noticed that fitlins has a --drift-model cosine option. Would this produce the same cosine regressors and be an appropriate substitute?


Just to be clear, the reason we insist on using the cosine basis with aCompCor is because the data is detrended prior to running CompCor, so any components found will be orthogonal to the cosine basis and will not include low-frequency components. Because you don’t have any cosine regressors, this requirement simply does not apply.

That said, yes, you can also attempt to directly low-pass filter in FitLins. The process for generating the cosine basis should be identical, so I wouldn’t be surprised if you only get a constant column for a short run.

1 Like