I am using fmriprep to preprocess some rs-fmri data and am using the acompcor regressors ( 5 WM and CSF) for confound regression.
However, while I was preparing the confound regressors, I realized the first two rows of the compcor regressors are all 0 (consisted across subjects).
When I check the residuals after regressing the confounds, the first two volumes of the residuals seems different than the rest of the volumes (more like the original functional image, less like the rest of the residuals, probably as a result of compcor regressors not being regressed from first two volumes).
Do you have any idea what may be the cause and how I can fix it?
Thanks for responding. Yes, it seems like each of them has been marked as outliers separately under non_steady_state_outlier00 non_steady_state_outlier01.
Reading previous threads, it seems like non_steady_outlier volumes impact the calculation of cosine regressors, but why also the compcor regressors? Is it because compcor regressors are calculated on the high pass filtered data (with cosine regressors)?
Also, is there a way to opt-out from getting non-steady-outliers calculated/being outputted? I havent explicitly define FD or DVARS thresholds but would something like setting both of FD and DVARS thresholds to 0 would get rid of it?
Thank you so much for the response. I just ran the data again and setting dummy scans explicitly to 0 solved the problem.
I have one more question. Since the acompcor regressors are derived from high-pass filtered data, if we are to filter the noise regressors prior to the de-nosing (so the frequencies of noise regressors and fMRI data match as mentioned in this paper, do you think the high-pass filtering should be omitted for acompcor regressors?
Thanks a lot again, for this and answering my other questions.