Advice on non-steady state regressors and other nuisance regs


I have a scan where the dummy volumes were already discarded - i.e. the data starts when the task actually starts (time 0 is already the first event in the trial). It’s a decision -making task - 12 runs of around 6-7 minutes each, TR=2.

In some of my runs, fmriPrep outputs different numbers of non-steady state regressors, usually in the first few volumes (even though this is already after the dummy volumes were discarded).

Should I use these non-steady state nuisance regressors, even though I know the dummy scans were already discarded and that this is already task data? And how does this choice influence which other nuisance regressors I should use?

Currently I’m planning to use the motion regressors, CSF CompCor regressors and Cosine regressors without using the non-steady state regressors. I can’t use the ICA-Aroma non-aggresively cleaned data because I want my data unsmoothed and in subject space.

I’d be very happy for any comments.