Slices semi-stripped in preprocessed dMRI

AmirHussein · December 18, 2025, 2:39pm

Summary of what happened:

Hi,

I’m using QSIprep 0.22.0 for preprocessing a dMRI dataset. For two subjects, the preprocessed data seem stripped -but not as the whole slices are missing- evident as early as in the denoising report:

The raw data has no evident issues. I repeated the analysis with QSIprep 1.0.2; the issue remained, but different slices were missing.

Command used (and if a helper script was used, a link to the helper script or the command generated):

Here is my QSIprep call:

singularity run --nv --cleanenv -B $TMPDIR:$TMPDIR \
qsiprep-v0.22.0.sif /input/ /output/derivatives/ participant \
--fs-license-file /toolboxes/license.txt \
--participant-label ${subjid} --output-resolution 1.25 \
--denoise-method dwidenoise --unringing-method mrdegibbs \
--hmc-model eddy --eddy-config /output/code/eddy_params.json --pepolar-method TOPUP \
--anat-modality T1w --anatomical-template MNI152NLin2009cAsym --b0_to_t1w_transform Rigid \
-w /user/jije2119/work --freesurfer-input /output/derivatives/fastsurfer_2.3.0 \
--stop-on-first-crash --nthreads 30 --skip_bids_validation

Here is the eddy_params.json file:

{
  "flm": "quadratic",
  "slm": "linear",
  "fep": false,
  "interp": "spline",
  "nvoxhp": 1000,
  "fudge_factor": 10,
  "dont_sep_offs_move": false,
  "dont_peas": false,
  "niter": 5,
  "method": "jac",
  "repol": true,
  "num_threads": 1,
  "is_shelled": true,
  "use_cuda": true,
  "cnr_maps": true,
  "residuals": true,
  "output_type": "NIFTI_GZ",
  "estimate_move_by_susceptibility": true,
  "mporder": 6,
  "args": "--ol_nstd=5"
}

Version:

0.22.0 and 1.0.2

Environment (Docker, Singularity / Apptainer, custom installation):

Singularity

Data formatted according to a validatable standard? Please provide the output of the validator:

I did not BIDS-validate the data, but 60 other subjects have been successfully preprocessed.

Relevant log outputs (up to 20 lines):

The preprocessing pipeline finishes normally without any error.

Screenshots / relevant information:

If necessary, I have 75 slices with a multi-band factor of 3. I did some additional debugging by looking in the work directory, and I’m not sure whether this is important, but I noticed that dwi_preproc_wf/hmc_sdc_wf/extract_b0_series/eddy_corrected_LPS_b0_series_mean.nii.gz has cuts (see below), whereas it isn’t in other subjects.

Any comments are appreciated.

Best,
Amir

Steven · December 19, 2025, 2:33pm

Hi @AmirHussein,

At first glance, not sure why this may happen. Any chance you can share a failing subject’s data?

Best,
Steven

AmirHussein · December 23, 2025, 10:53am

Hi Steven,

Thanks for your reply. I just shared the file with you via email.

The entire dataset I’ve worked with contains only dMRI data acquired with reverse PE b0, except for three subjects. For these three, I used Synb0 following Ingress of Synb0-DISCO corrected data into qsiprep · PennLINC/qsiprep · Discussion #583 · GitHub, and reran QSIprep. Thus, acq-dmri_dir-PA_epi.nii.gz files are actually generated by Synb0. I think this discussion may also be helpful if there are any plans to integrate the Synb0-DISCO pipeline into QSIprep.

The issue I mentioned occurred for two of the three subjects; one completed successfully without those stripes.

I took additional steps to identify the source of the issue. When I ran QSIprep with --ignore fieldmaps, the stripes disappeared, and the results returned to normal (with no SDC, of course). I also noticed that, please correct me if I’m wrong, QSIprep concatenates everything together before running denoising. If there are any excessive movements between the concatenated images, the denoising will not succeed properly: Preprocessing — qsiprep 1.0.2 documentation and Preprocessing — qsiprep 1.0.2 documentation.

Comparing the framewise-displacement diagrams of the with and without --ignore fieldmaps is shown below, respectively:

High FD at the beginning (max = 41) led me to believe that the merging and denoising were the issue, so before QSIprep, I linearly registered the RPE b0 to the main dMRI sequence with dof = 6. This time, the results were much better (compare them to the one I shared at the beginning of the discussion), but the issue persisted:

Looking forward to your comments.

Best,
Amir