WM, CSF and aCompCor confounds output from fmriprep


I have short question regarding WM, CSF and aCompCor confounds from fmriprep .tsv file. Are these confounds calculated within eroded or complete WM/CSF masks?



The eroded union of CSF and white matter masks. Please see the reports for outlines of the ROIs for each BOLD series.


The relevant code is here: https://github.com/poldracklab/fmriprep/blob/master/fmriprep/workflows/bold/confounds.py#L148

1 Like


I wanted to follow up on this question.

We are planning to do WM and CSF regression post-ICA Aroma. Thus, we plan to extract WM and CSF time series ourselves and recalculate.

I assume we would use the output: sub-001_space-MNI152NLin6Asym_label-CSF_probseg.nii if we want a CSF mask in MNI space or sub-001_label-CSF_probseg.nii if we wanted it in T1 native space. And similarly for the WM masks.

Is this correct? However are these masks already eroded or would we need to do the erosion step if we wanted to re-calculate WM and CSF signals?

Thanks so much!


1 Like

I have exactly the same question, did you figure this out?