Preprocessing MRI data by fmriprep amd Tedana

, , ,
1

Hi

I am working on multi-echo data and am a bit confused about the steps after tedana. We performed slice-time correction, head-motion correction, and susceptibility-distortion correction before running tedana. After tedana, I plan to rerun fMRIPrep while skipping slice-time correction and susceptibility-distortion correction. I would also like to skip head-motion correction, but I’m not sure how to do that in fMRIPrep. Thank you!

Best,
Gucy

2

What is your goal in rerunning fMRIPrep?

In order to denoise data in target spaces, we recommend either applying transforms from fMRIPrep to the boldref-space data or, more easily, using the noise components flagged by tedana to denoise the fMRIPrep-generated data in whatever space you want.

3

Hi,

The ‘–me-output-echos’ generated by fMRIPrep are in native space, so the denoised images produced by tedana remain in native space as well. My goal in the next step is to normalize these denoised images to standard space. I currently have three types of transformation files: from-boldref_to-T1w_mode-image_desc-coreg_xfm.txt, from-orig_to-boldref_mode-image_desc-hmc_xfm.txt, and from-T1w_to-MNI152NLin2009cAsym_mode-image_xfm.h5. May I confirm whether the following code is correct if I only want to normalize the denoised images to standard space?

*antsApplyTransforms *

  • -d 3 -e 3 *
  • -i denoised_bold_in_origspace.nii.gz *
  • -r $FMRIPREP_DIR/anat/sub-XXX_space-MNI152NLin2009cAsym_desc-preproc_T1w.nii.gz *
  • -t from-T1w_to-MNI152NLin2009cAsym_mode-image_xfm.h5 *
  • -n Lanczos *
  • -o denoised_bold_space-MNI152NLin2009cAsym.nii.gz*

Thank you for your explanation!

Best,
Gucy

4

The echo-wise images produced by fMRIPrep will be in boldref space (i.e., after slice timing correction, motion correction, and susceptibility distortion correction), rather than orig space (the space of the raw data before any preprocessing), so if you want to denoise those files, you will need to use both the T1w-to-MNI transform and the boldref-to-T1w transform.

I’m more familiar with the antspyx interface than command-line ANTs, but, assuming the other parameters are correct, you would need to add the boldref-to-T1w transform to the call, like below:

antsApplyTransform \
    -d 3 \
    -e 3 \
    -i denoised_bold_in_origspace.nii.gz \
    -r $FMRIPREP_DIR/anat/sub-XXX_space-MNI152NLin2009cAsym_desc-preproc_T1w.nii.gz \
    -t from-T1w_to-MNI152NLin2009cAsym_mode-image_xfm.h5 from-boldref_to-T1w_mode-image_desc-coreg_xfm.txt \
    -n Lanczos \
    -o space-MNI152NLin2009cAsym_desc-denoised_bold.nii.gz
5

Hi Taylor,

Thank you so much for your clarification. This has been very helpful.

I’ve reviewed a few documents describing workflows that combine fMRIPrep and tedana, and I’d like to confirm whether the following steps seem reasonable to you:

  1. Generate echo-wise images with fMRIPrep.
  2. Apply tedana to obtain denoised outputs.
  3. Normalize tedana-generated images to the target space(s).
  4. Regress out head-motion effects by including the motion covariates generated by fMRIPrep (Step 1) in the first-level model.

In addition, I have drafted the following command for applying the transforms. Could you let me know if this looks appropriate? Thank you!

bold = ants.image_read("denoised_tedana.nii.gz")
ref = ants.image_read(
    "$FMRIPREP_DIR/anat/sub-XXX_space-MNI152NLin2009cAsym_desc-preproc_T1w.nii.gz"
)
t1w_to_mni = "from-T1w_to-MNI152NLin2009cAsym_mode-image_xfm.h5"
boldref_to_t1w = "from-boldref_to-T1w_mode-image_desc-coreg_xfm.txt"
bold_mni = ants.apply_transforms(
    fixed=ref,
    moving=bold,
    transformlist=[t1w_to_mni, boldref_to_t1w],
    interpolator="lanczosWindowedSinc",
    imagetype=3  
)
ants.image_write(bold_mni, "denoised_bold_space-MNI152NLin2009cAsym.nii.gz")

Best,
Gucy