I am working with resting-state data from ~200 subjects.
After preprocessing it in fMRIprep, I want to apply smoothing, then denoising in CONN toolbox.
What are the recommended confounds to include from fMRI-prep for the denoising step beyond the 6 motion parameters and their derivates, powers, etc.?
I found several recommendations for adding a_Com_corr_xx – but I am wondering how much agreement there is about including them.
Would a specific cut-off based on the variance explained by each of them be useful when selecting the number of a_Com_corr_xx to include? Or is it more beneficial to go with the full set?
I expect that the number of anatomical and temporal Comp_corr will be different for each subject so keeping a similar portion of explained variance, seem like a solution?
Are there any benefits of adding t_Como_corr_xx parameters as well?
Are there other parameters important to add when dealing with resting-state data?
Depending on how much work you want to do in terms of migrating fMRIPrep outputs to CONN, the latest releases of CONN currently support a direct import from an fMRIPrep directory. The recommendation is to run smoothing before proceeding to denoising. In the denoising, CONN will “automatically” select the confounds from the fMRIPrep output that match the standard CONN pipeline. This includes re-computing the CONN version of CompCor and using the first 5 components from the WM and CSF.
Beyond that, you can obviously select any confound regressors you want from the fMRIPrep outputs. Consider looking at this article as a guide to the relative performance of regressor selection.
Be aware that if you choose the pre-computed derivatives or squared derivatives of any regressor, CONN will throw an error unless you change the NaN in the first row.
Hello Adam, could I trouble you with one more question? By re-computing the CONN versions of CompCor do you mean reformatting those included in the confounds file from fMRIprep, or calculating them anew?
And if the later – is there a way to remove this step from CONN and rely on those from fMRIprep?
Another question is about NaN in the first raw. As of now it seems that CONN will read the regressors file only if NaN is replaced by a numeric value, so not space or dot. In this case is 0 or 1 a better choice? I mean even if to avoid imaginary numbers.
Regarding NaN – a q&d 1-liner approach is to simply replace the 1st by the 2nd row (i.e., 2nd by 3rd bc of of heading line) to keep up with all kind of different value scales from large & diverse fmriprep regressor files.
The 1-liner (bash): line3=$(sed ‘3q;d’ $regfile); sed “2s/.*/${line3}/” $regfile > $regfile_noNaN