CSF and WM regression

[cchang]

Hi all,

I’m using fmriprep for fmri preprocessing and then using the confounds_regressors.tsv file to regress out CSF + WM signals. A lot of papers have mentioned using 6 acompcor vectors. However, it seems like some people use percentages (at least 75% of variance explained in the acompcor). To get 75% variance, I would be including 120 acompcor vectors for regression.

How are people approaching this?

[Steven]

fMRIPrep Confounds for GLM

Hello all, I am trying to use some of the confounds generated by fMRIPrep to denoise “task” data within the FSL FEAT GLM, and I am wondering which columns to include as regressors. I’ve followed along with this previous post so far, but wanted to ask for some further instruction/validation of methods. I am planning to include the 6 major motion regressors (x,y,z trans & rot) and some combination of a/c/tCompCor regressors to account for physiological noise, but my question lies in which of th…

[cchang]

Gotcha. Thanks @Steven !

A quick follow up question. So, if fmriPrep says 122 components explain 50% of my variance, then it would make sense to include all 122 components from the _confounds.tsv file in my physiological regression?

Why is it that some people only have 5 or 6 components vs. my 122 components?

[Steven]

Hi @cchang,

@cchang

So, if fmriPrep says 122 components explain 50% of my variance, then it would make sense to include all 122 components from the _confounds.tsv file in my physiological regression?

If you want to be consistent in regressing out 50% variance, then yes.

@cchang

Why is it that some people only have 5 or 6 components

Others choose to prioritize keeping temporal degrees of freedom (tDOF) in the GLM the same across subjects, which (assuming the scan lengths are constant) would mean including the same amount of regressors in each model. Especially in shorter scans, removing too many tDOF could lead to inflexible model fitting.

Best,
Steven

[cchang]

@Steven

Others choose to prioritize keeping temporal degrees of freedom (tDOF) in the GLM the same across subjects, which (assuming the scan lengths are constant) would mean including the same amount of regressors in each model. Especially in shorter scans, removing too many tDOF could lead to inflexible model fitting.

This makes a lot of sense. My runs are currently around 10 mins each. What would be considered as a “shorter scan”? As for the tDOF, does this mean that they’re extracting 5 or 6 components from the tCompcor or the aCompcor?

[Steven]

Hi @cchang,

@cchang

As for the tDOF, does this mean that they’re extracting 5 or 6 components from the tCompcor or the aCompcor?

aCompcor

@cchang

This makes a lot of sense. My runs are currently around 10 mins each. What would be considered as a “shorter scan”?

Depends on the TR, but that seems okay. You can always test different schemes on held out subjects if they are not going to be included in your analysis.

Best,
Steven

[cchang]

Gotcha - Thanks so much!