eddy (FSL)

qsiprep.workflows.dwi.fsl.init_fsl_hmc_wf()

--hmc-method eddy (the default) runs FSL’s eddy [1], which corrects head motion, eddy-current distortion and, when it is given a field, susceptibility distortion in one model. It requires a shelled sampling scheme.

What runs

1. The concatenated series is reoriented to LAS+, which the FSL tools expect.

2. If there is reverse phase-encoded data, TOPUP [2] estimates the susceptibility field from b=0 images (see below).

3. eddy estimates and corrects head motion and eddy currents, and applies the TOPUP field or a GRE fieldmap in the same resampling. Its outlier replacement [3] is on by default.

4. The corrected series is converted back to LPS+. Any remaining distortion correction (DRBUDDI, SyN) is applied to these outputs.

eddy writes its own corrected volumes, so the output data are interpolated twice: once by eddy and once by QSIPrep’s final resampling into ACPC space. eddy also applies Jacobian intensity modulation to its own corrections whenever its resampling method is jac (the default); see Intensity modulation.

The motion parameters, the eddy-current coefficients and the outlier counts per slice go to the confounds file and the carpet plot. The contrast-to-noise ratio maps eddy computes per shell are written as stat-cnr derivatives.

Distortion correction cases

No fieldmap. Only head motion and eddy currents are corrected.

(Source code, png, svg, pdf)

Reverse phase encoding (TOPUP). b=0 images are selected from each distortion group of the correction unit, whether they come from the DWI series or from epi fieldmaps; images native to the series being corrected are preferred over borrowed ones. TOPUP estimates the field from them, and eddy applies it with --topup. With --sdc-method topup+drbuddi, DRBUDDI then refines eddy’s output; this needs a single matched pair of reverse phase-encoded DWI series, since with a lone reverse b=0 the output is already unwarped and a second pass would correct it twice. With --sdc-method drbuddi, TOPUP is skipped and DRBUDDI corrects the eddy output on its own.

(Source code, png, svg, pdf)

GRE fieldmap. The fieldmap is converted to Hz, rigidly aligned to eddy’s first volume, and passed to eddy with --field, which applies it inside its model the way it applies a TOPUP field. Setting "estimate_move_by_susceptibility" in --eddy-config additionally lets eddy estimate how the field changes with head position [4]. Before 26.1 the fieldmap was applied to eddy’s outputs instead; --force gre-sdc-after-eddy restores that, for comparison, and is deprecated.

(Source code, png, svg, pdf)

SynB0. With --sdc-anat-reference synb0, the synthetic distortion-free b=0 (see SynB0) joins the TOPUP inputs as a volume with zero readout time, and eddy uses the resulting field like any other. This needs topup in --sdc-method.

SyN. With --sdc-anat-reference invt1w, the SyN correction is estimated and applied on eddy’s outputs.

--sdc-anat-reference t2w has no effect with eddy; T2Wreg is a TORTOISE tool.

Configuring eddy

eddy has many options. Rather than exposing them on the command line, QSIPrep reads them from a JSON file given with --eddy-config. The default file is eddy_params.json; the keys are eddy’s own option names. The file is validated when the command line is parsed. --gpu eddy runs eddy_cuda and overrides the file’s use_cuda.

References

[1]

Jesper L.R. Andersson and Stamatios N. Sotiropoulos. An integrated approach to correction for off-resonance effects and subject movement in diffusion mr imaging. NeuroImage, 125:1063–1078, 2016. doi:10.1016/j.neuroimage.2015.10.019.

[2]

Jesper L.R. Andersson, Stefan Skare, and John Ashburner. How to correct susceptibility distortions in spin-echo echo-planar images: application to diffusion tensor imaging. NeuroImage, 20(2):870–888, 2003. doi:10.1016/S1053-8119(03)00336-7.

[3]

Jesper L. R Andersson, Mark S. Graham, Enikő Zsoldos, and Stamatios N. Sotiropoulos. Incorporating outlier detection and replacement into a non-parametric framework for movement and distortion correction of diffusion mr images. NeuroImage, 141:556–572, 2016. doi:10.1016/j.neuroimage.2016.06.058.

[4]

Jesper L.R. Andersson, Mark S. Graham, Ivana Drobnjak, Hui Zhang, and Jon Campbell. Susceptibility-induced distortion that varies due to motion: correction in diffusion mr without acquiring additional data. NeuroImage, 171:277–295, 2018. doi:10.1016/j.neuroimage.2017.12.040.