Integrating fmriprep & CONN: how to handle denoising step?

1

Hi all,

I’ve done a bunch of reading through the archives both here and on the CONN listserv about denoising, but I’m still unclear about how to best use fmriprep-processed ICA-AROMA (non-aggressive) files in CONN (18b).

  • At the denoising step, I would normally include WM, CSF, effect of rest, and ART regressors. I have gathered that I probably should not include motion regressors here if I am using non-aggressive AROMA data. But should I still include WM, CSF, and effect of rest as confounds at the denoising step? Or have those effects already been removed?

  • Also at the denoising step: include detrending, or no?

  • Finally, on the fact that CONN and fmriprep are using different versions of MNI space: transform the images, transform the CONN atlas, ignore the issue because the effect is likely tiny, or some other solution?

Thanks in advance for any advice.

2

Hi, I can reply for the first part of the question maybe. From my understanding, ICA AROMA was trained to classify motion-related noise, not physiological noise (i.e., WM, CSF). So, if you want, there is a rational for including them as additional confounders. That said, it has been showed that ICA AROMA performs pretty well regardless of whether they are incorporated (e.g., see Parkes et al., 2018 Neuroimage).