Quick Start

There are many options for running qsiprep but most have sensible defaults and don’t need to be changed. This page describes the options most likely to be needed to be adjusted for your specific data. Suppose the following data is available in the BIDS input:

sub-1/ses-1/dwi/sub-1_ses-1_acq-multishell_run-01_dwi.nii.gz
sub-1/ses-1/dwi/sub-1_ses-1_acq-multishell_run-01_dwi.nii.gz
sub-1/ses-1/dwi/sub-1_ses-1_acq-multishell_run-03_dwi.nii.gz
sub-1/ses-1/fmap/sub-1_ses-1_dir-PA_epi.nii.gz

One way to process these data would be to call qsiprep like this:

qsiprep \
  /path/to/inputs /path/to/outputs participant \
  --output-resolution 1.2 \
  --fs-license-file /path/to/license.txt

Grouping scans

Note

This section explains --separate-all-dwis, --denoise-after-combining and --dwi-denoise-window

Assuming that sub-1/ses-1/fmap/sub-1_dir-PA_epi.nii.gz has a JSON sidecar containing the IntendedFor field for fieldmap correction (see here):

"IntendedFor": [
  "ses-1/dwi/sub-1_ses-1_acq-multishell_run-01_dwi.nii.gz",
  "ses-1/dwi/sub-1_ses-1_acq-multishell_run-02_dwi.nii.gz",
  "ses-1/dwi/sub-1_ses-1_acq-multishell_run-03_dwi.nii.gz"
]

qsiprep will infer that the dwi scans are in the same warped space - that their susceptibility distortions are shared and they can be combined before head motion correction. Since we didn’t specify --separate-all-dwis the separate scans will be merged together before head motion correction and the fully preprocessed outputs will be written to derivitaves/qsiprep/sub-1/ses-1/dwi/sub-1_ses-1_acq-multishell_desc-preproc_dwi.nii.gz. otherwise there will be one output in the derivatives directory for each input image in the bids input directory.

It is beneficial to have as much data as possible available for head motion correction. However, the denoising preprocessing step has important caveats that should be considered. For a discussion see Denoising and Merging Images.

Specifying outputs

Note

This section covers --output-resolution 1.2, and --skip-t1-based-spatial-normalization.

Unlike with fMRI, which can be coregistered to a T1w image and warped to a template using the T1w image’s spatial normalization, the T1w images do not contain enough contrast to accurately align white matter structures to a template. For this reason, spatial normalization is typically done after models are fit. Therefore we omit the --output-spaces argument from preprocessing: i.e. no template warping takes place by default.

All outputs will be registered to the T1w image (or the AC-PC aligned b=0 template if --dwi-only was specified) but will have an isotropic voxel size. Furthermore, all outputs are aligned according to the AC-PC convention: the coordinates are changed from the native scanner coordinates to a new system where $0, 0, 0$ is where the midline intersects the anterior commissure (AC).

Cortex can be accurately spatially-normalized using the T1w image, so the T1w image is still spatially normalized by default during preprocessing. The transform from the T1w image to the MNI152NLin2009cAsym template is included in the derivatives. This can be used during reconstruction to map cortical parcellations from the template into the DWI in order to estimate brain graphs. If you want to save ~20 minutes of computation time, this normalization can be disabled with the --skip-t1-based-spatial-normalization option.

The --output-resolution argument determines the spatial resolution of the preprocessed dwi series. You can specify the resolution of the original data or choose to upsample the dwi to a higher spatial resolution. Some post-processing pipelines such as fixel-based analysis recommend resampling your output to at least 1.3mm resolution. By choosing this resolution here, it means your data will only be interpolated once: head motion correction, susceptibility distortion correction, coregistration and upsampling will be done in a single step. If your are upsampling your data by more than 10%, QSIPrep will use BSpline interpolation instead of Lanczos windowed sinc interpolation.

Head motion correction model

Although FSL’s eddy is technically model-free, it is an option for --hmc-model along with 3dSHORE and none. Choosing eddy (the default) runs FSL’s eddy for head motion correction and eddy current correction. This will work for single-shell and multi-shell sampling schemes. The 3dSHORE (aka “SHORELine”) option works for multi-shell, Cartesian grid sampling (DSI) and random q-space sampling (CS-DSI).

The option none will register all the b=0 images to one another and the b>0 images will have the transform from the nearest b=0 image applied. This is not recommended. Between eddy and 3dSHORE, all sampling schemes can be motion corrected.

Enabling and disabling preprocessing steps

The image processing operations performed by QSIPrep are configured by default to apply to most generic sequences. Depending on your sequence and sampling scheme, you can elect to enable, disable or alter the behavior of these steps to better match your data.

	Denoising	Gibbs Unringing	B1 Bias Field Correction
Description	Reduce random noise in images.	Remove spatial ringing artifact from images.	Remove spatial non- uniformity of images.
Algorithms	`dwidenoise` (MRtrix3) patch2self (DIPY)	`mrdegibbs` (MRtrix3)	`dwibiascorrect` (ANTs/MRtrix3)
Default	`dwidenoise` (MRtrix3)	None applied	`dwibiascorrect` (ANTs/MRtrix3)
Disable with	`--denoise-method none`	Disabled by default	`--dwi-no-biascorr`
Change behavior with	`--dwi-denoise-window N` changes denoising window to N voxels	`--unringing-method` enables Gibbs unringing	No parameters
Notes	Set the window to `auto` or a specific voxel number	Technically only supposed to be run on full Fourier acquisitions.	Uses N4BiasFieldCorrection on b=0 images, applies correction to the whole series

Not included in this table is the b=0 intensity harmonization step, which applies simple scaling if there is more than one NIfTI file being processed. It can be disabled with --no-b0-harmonization.

Each of these steps can be applied at the same time, which by default is before any images are concatenated. The user can instead run these steps together after images are concatenated by specifying --denoise-after-combining. See Denoising and Merging Images for more info.

What is happening??

While QSIPrep runs with -v -v, you will see lots of unintuitive output in the terminal like:

[Node] Setting-up "qsiprep_wf.single_subject_PNC_wf.dwi_finalize_acq_realistic_wf.transform_dwis_t1.final_b0_ref.b0ref_reportlet" in "/scratch/qsiprep_wf/single_subject_PNC_wf/dwi_finalize_acq_realistic_wf/transform_dwis_t1/final_b0_ref/b0ref_reportlet".
  201229-21:33:46,213 nipype.workflow INFO:
    [Node] Running "b0ref_reportlet" ("qsiprep.niworkflows.interfaces.registration.SimpleBeforeAfterRPT")
  201229-21:33:48,51 nipype.workflow INFO:
    [MultiProc] Running 2 tasks, and 3 jobs ready. Free memory (GB): 3.70/4.00, Free processors: 0/2.
                      Currently running:
                        * qsiprep_wf.single_subject_PNC_wf.dwi_finalize_acq_realistic_wf.transform_dwis_t1.final_b0_ref.b0ref_reportlet
                        * qsiprep_wf.single_subject_PNC_wf.anat_preproc_wf.mni_mask

These print-outs describe what is currently running. In this case, we can see that b0ref_reportlet and mni_mask are being run simultaneously. What exactly are these steps and how do they fit into the overall workflow?

We can find the name of the node (aka “job”) being run in the quotation marks. This task can be found in the workflow diagrams in Preprocessing pipeline details. In the case of mni_mask it is part of Handling Lesions and abnormalities, while b0ref_reportlet is part of DWI reference image estimation. The relative place of these jobs’ parent workflows in the overall workflow can be seen in the graph of DWI preprocessing.

Also in this example you can see that QSIPrep was run with --nthreads 2 (Free processors: 0/2) and that both open slots are running a job.