help
help > RE: fmriprep to conn approach
Jul 28, 2021 10:07 AM | Alfonso Nieto-Castanon - Boston University
RE: fmriprep to conn approach
Hi Walker,
Yes, that sounds perfectly fine and I do not foresee any potential issues/problems with that approach (the only issue that comes to mind is that CONN's white/CSF ROIs in this scenario are going to be unusual, as the default behavior is that they are estimated in a subspace orthogonal to all of your specified covariates which in this case include already-computed aCompCor components; but in any case this will not have any effect in your procedure as you will simply be disregarding these white/CSF ROIs)
Best
Alfonso
Originally posted by Walker Pedersen:
Yes, that sounds perfectly fine and I do not foresee any potential issues/problems with that approach (the only issue that comes to mind is that CONN's white/CSF ROIs in this scenario are going to be unusual, as the default behavior is that they are estimated in a subspace orthogonal to all of your specified covariates which in this case include already-computed aCompCor components; but in any case this will not have any effect in your procedure as you will simply be disregarding these white/CSF ROIs)
Best
Alfonso
Originally posted by Walker Pedersen:
Hi,
I'm working on a pipeline for taking resting state data through fmriprep and then into CONN for smoothing and denoising. Rather than importing as an fmriprep file, I've found it easier to just create a tsv file with all of the confound variables (output by fmriprep) I want to include in denoising for a given run, entering it as a first-level covariate during setup and adding it as a confound in denoising, with no other confounds. (I also have detrending and bandpass filtering turned on during denoising.)
The confound file I've constructed has motion parameters, 10 components of acompcorr, 3 regressors for the first 3 TRs to remove pre-steady state noise, and a regressor for each TR above my FD motion cutoff.
Does this approach make sense? It seems pretty straight forward, but since I am straying from the standard CONN pipeline a bit, I want to make sure I'm not overlooking something that would make this approach not work as intended.
An example of one of the confound files is attached if that helps.
Thanks!
I'm working on a pipeline for taking resting state data through fmriprep and then into CONN for smoothing and denoising. Rather than importing as an fmriprep file, I've found it easier to just create a tsv file with all of the confound variables (output by fmriprep) I want to include in denoising for a given run, entering it as a first-level covariate during setup and adding it as a confound in denoising, with no other confounds. (I also have detrending and bandpass filtering turned on during denoising.)
The confound file I've constructed has motion parameters, 10 components of acompcorr, 3 regressors for the first 3 TRs to remove pre-steady state noise, and a regressor for each TR above my FD motion cutoff.
Does this approach make sense? It seems pretty straight forward, but since I am straying from the standard CONN pipeline a bit, I want to make sure I'm not overlooking something that would make this approach not work as intended.
An example of one of the confound files is attached if that helps.
Thanks!
Threaded View
| Title | Author | Date |
|---|---|---|
| Walker Pedersen | Jul 27, 2021 | |
| Alfonso Nieto-Castanon | Jul 28, 2021 | |