SHORELine

qsiprep.workflows.dwi.hmc_sdc.init_qsiprep_hmcsdc_wf()

Warning

SHORELine is scheduled for removal in a future major release. Use --hmc-method tortoise for non-shelled data; it corrects eddy currents as well as head motion, which SHORELine does not.

--hmc-method shoreline runs QSIPrep’s own model-based head motion correction [1], written for DSI and other schemes that eddy cannot handle. It corrects head motion only.

What runs

  1. All b=0 images are aligned to a mid-point b=0, and each b>0 image gets the transform of its nearest b=0.

  2. For each b>0 image, a 3dSHORE [2] (or tensor) model is fit to all the other images, and used to predict the signal at the left-out image’s q-space coordinate. The image is registered to that prediction and its gradient vector rotated accordingly.

  3. A new model is fit to the updated images and vectors, and the leave-one-out registration is repeated ("iters" times, default 2).

  4. Susceptibility distortion correction runs on the result: DRBUDDI for reverse phase-encoded data, or a GRE fieldmap or SyN. synb0 and t2w anatomical references are not supported on this backend.

The motion transforms are carried, not applied, so the data are interpolated once, at the final resampling. Six (or twelve, with an affine transform) parameters per volume go to the confounds file, and the per-slice model fit is shown in the carpet plot.

../_images/hmc_shoreline-1.png

(Source code, png, svg, pdf)

Configuring SHORELine

--shoreline-config is a JSON file; the default is shoreline_params.json. Every key is optional; unknown keys are an error.

Key

Allowed values

Default

Meaning

model

"3dshore", "tensor", "none"

"3dshore"

Signal model used to predict each left-out image. "none" skips the model and gives each b>0 image the transform of its nearest b=0 image, which is not recommended.

iters

An integer of at least 1

2

Number of SHORELine iterations.

transform

"Affine", "Rigid"

"Affine"

Transformation optimized during head motion correction.

References