Glad to find this discussion here. Recently I worked on a analysis pipeline using FMRIPREP. And here’s what I did in preprocess after FMRIPREP for later modelling.
- Grab *bold_space-MNI152NLin2009cAsym_preproc.nii.gz outputs and normalized its median intensity to 10000 (follow FSL’s convention) .
- Use SUSAN smoothed the data with x mm FWHM.
- Optional Applied non-aggressive ICA-AROMA denoise as FMRIPREP document mentioned.
- Optional Use fsl_glm to regress out nuisance like WM from _confounds.tsv and taken the residuals images.
- Applied temporal filtering use fslmath -bptf (I think if you applied a highpass filter, it will taken care of the linear detrend.)
- Add back the mean image derived from 2 or 3 to make the filtered image a firm dataset.
I have several question regrading to this pipeline.
- As you mentioned,
Create a denoising design matrix that would include the following columns from the _confounds.tsv file: AROMAAggrComp* (those are the noise components from ICA AROMA), WhiteMatter, and aCompCor01 (first component of the anatomical compcor is a good approximation of the mean signal in WM and CSF). In addition to achieve the linear detrending you should add a column of ascending numbers (1,2,3,4 etc).
should I combined 3 and 4 into one step use fsl_glm? Is that an aggressive way to remove nuisance variables?
I’m aware that the in FMRIPREP, ICA-AROMA was calculated on smoothed data with 6mm FWHM. Could I apply the derived regressors on smoothed data with different FWHM (or unsmoothed data for MVPA analysis)?
Could I use these ICA-AROMA regressors on data in T1w space? Or some modification are needed?
Another question not quite related to this thread. If it’s inappropriate here I could open a new thread later.
I obtained several 7T MRI data as a benchmark for later study. For structure image, I got a MP2RAGE image with 0.7 mm resolution. After some tweaks, I still couldn’t get a good skullstrip result from FMRIPREP due to the image contains noise voxels around the brain with similar intensity. Is it possible to modifies anat workflow to handle MP2RAGE image?
Also, the partial FOV bold image corregistration was failed in this dataset (maybe due to problematic anat image). Should I scan a whole brain reference EPI image with same parameters (_sbref ?) to help registration? Could FMRIPREP support it?