Location of bold reference image in fmriprep 1.0.6?

fmriprep

#1

Hi, all. We’d like to do some analyses on the bold reference images across our studies (to see if we can classify the originating scanner), but I don’t see that the reference images are output.

I think I should be able to fetch them from the working directory, but I wanted to confirm the correct location. If I use:
workingdir/fmriprep/fmriprep_wf/single_subject_sub_wf/func_task_run_001_wf/bold_reference_wf/gen_ref/ref_image.nii.gz

will I be getting the reference image I’m looking for?

Thanks!
Todd


#2

I think this is it, but we should probably output those with other results.


#3

That is the initial reference image, but some additional steps are done to improve the gray/white contrast (for boundary-based registration).

The enhanced image is: bold_reference_wf/enhance_and_skullstrip_bold_wf/n4_correct/*.nii.gz
The skull-stripped enhanced image is: bold_reference_wf/enhance_and_skullstrip_bold_wf/apply_mask/*.nii.gz

Those are the images actually used for further processing.


#4

Thanks, Chris(es),

My RA pointed out to me that we actually want the reference image in MNI space to run the classification algorithm on.

Seems like I might want:
/fmriprep_wf/single_subject_sub_wf/func_preproc_task_task_run_001_wf/ bold_reference_wf/enhance_and_skullstrip_bold_wf/apply_mask/uni_xform_masked.nii.gz

Is that correct?

Thanks so much, again,
Todd


#5

I don’t think we actually apply the full set of transforms to the reference image so this file is not available.

Why not take an average of 10 first volumes of your _preproc outputs in the MNI space?


#6

My backup plan was to take the overall average of the timecourse, but I was thinking that the way fmriprep identifies the representative image might be slightly better than a simple average.

Is there any reason that the average of the first 10 images would be better than a whole timecourse average? (This is a block design experiment, so there are going to be time-varying differences in some ROIs.)


#7

You want to average over some TRs (so you would remove noise), but you expect variance over a large number of TRs (so it’s a trade off). I would also skip some TRs in the beginning to avoid catching the non-steady state effects.