Wm and csf confounds

Hi everyone,

I was wondering if it is all right to use the average WM and CSF signal in the *desc-confounds_timeseries.tsv file for denoising post-ICA AROMA instead of using the CompCor outputs.

Thanks!
Best
Hans

Hi Hans,

The short (if a bit unsatisfying answer) story suggested across a wide range of fMRI denoising papers seems to be that there is no one size fits all approach that works best. However, I think acompcor components tends to remove physiological noise better than average WM/CSF signal. At the same time, the higher number components may tend to remove temporal degrees of freedom, so sometimes (e.g. scans with few volumes), you may opt to use fewer acompcor components or a different strategy.

What kind of data do you have and what kind of analyses are you going to run with it?

Best,
Steven

Dear Steven,

Thanks for you swift response.
I have resting-state fMRI data, and I’m performing spatially constrained ICA after preprocessing, followed by dyn FNC analysis.
At first I was thinking of only running ICA AROMA, without additional WM and CSF regression. Since I’m conducting spatially constrained ICA with only neural template components, I wondered whether additional regression might be worthwhile. I know that Group ICA + backreconstruction can pull out CSF and WM components quite well. What are your thoughts about this?

Thanks again.

Best
Hans

Sorry, I am not familiar with scICA or back reconstruction. Hopefully someone else can help!

Okido, thanks for your help so far.