Hi everyone,

I hope someone can help me with two issues regarding NSR in C-PAC.

(a) The first once concerns the order of the NSR steps implemented in the default C-PAC pipeline. I think all the desired steps (e.g. aCompCor, motion correction, GSR etc.) are performed simultaneously and not in a modular/step-wise fashion but can anyone confirm this is correct? An additional step that can be performed after or before regression is the band-pass filtering.

(b) The second issue concerns global signal regression. As CSF and WM contain some variance related to GSR, it appears possible that CompCor may actually account for some correction similar to GSR. This would of course be related to how the CSF and WM look like, and it seems that the default pipeline includes erosion (thus minimizing a possible effect). But it's unclear to me what the level of erosion is (e.g., 30mm?). For instance, there is some evidence that eroding the WM twice and CSF four times removes the risk of unwilling correcting for global signal (Power et al., 2017: https://doi.org/10.1016/j.neuroimage.201...). I'm trying to avoid the possibility of regressing the global signal "twice", once via CompCor and once via GSR, as I have both active in my pipeline. Is there a risk of this? The CSF and WM masks used in the NSR also appear to have a high threshold (0.95), so I am uncertain whether additional erosion procedures (i.e., in addition to the default) are necessary. I am also uncertain how the default erosion (30mm) corresponds to masks twice or four times eroded, like mentioned in the Power paper referenced above.

I'd really appreciate your help!


Cheers,
Mihai