How to evaluate if DWI susceptibility-induced distortion correction is effective for T1-to-dwi registration?

Here (Anatomically-Constrained Tractography (ACT) — MRtrix 3.0 documentation) it is mentioned that the rigid body registration of T1 images to DWI (required for ACT) requires an effective distortion correction. This includes an effective susceptibility-induced distortion correction.

My question is how to evaluate the effectiveness of the susceptibility-induced distortion correction? It is easy to visually compare the outline of the registered-T1 and original dwi images to see if they fit, but what is the best way to see if the inner brain structures are aligned? Is there some metric I can use? If visual examination is my best resort, what parts of the brain I should examine most closely?
Alternatively, if examining alignment of the outline of the T1 and dwi is enough, please let me know. I assume that there are cases when the outlines fit but the internal structures not, but maybe it’s very unlikely (I’m especially interested in the case of distortion correction with topup – is it possible that after topup, the dwi and T1 outlines will align well, but the inner structures of the brain wouldn’t?).

Alternatively, I can also examine the results of the susceptibility-induced distortion correction independently of the T1. I guess that visual inspection is inherently limited because we don’t really have a ground-truth of how the corrected dwi should look like. However, if I’ve used topup, so I know that eddyQC provides some metrics on the performance of topup that I can look at. That said, I don’t have any reference on what is considered acceptable for proper T1 to dwi registration, and what should be considered an outlier? Can you provide any guidelines?


1 Like

Dear @orencivier ,

I have found the last pages of this resource very helpful for understanding this topic.

There is some subjectivity in determining whether the correction is adequate or not. If there are objective tools to quantify this, I would be interested as well!


Thank you @NicDC. In the resource you sent, they generate the outline of the white matter based on the T1 and then overlay it on the EPI image (fMRI in their case, but can be dwi for my needs). That sounds like a neat idea of how to visually evaluate a good match.
Unfortunately, however, it is not clear how they generated the outline (I believe they used FSL’s FEAT, but not sure).

Does anybody know how I can generate the white matter outline based on T1 using MRtrix3 tools? I’m aware of the 5tt2gmwmi command, but this is limited to the border between WM and GM. I’m also interested in the border between WM and CSF.


Yes indeed, FEAT is used to get the WM segmentation and then the boundary voxels are obtained.

To get the WM segmentation you will anyway use FEAT or FreeSurfer (as this cannot be done with MRtrix at the moment).

The steps to then get the outline are:

  1. binarize the WM mask

mrcalc wm.nii.gz 0 -gt wm_bin.nii.gz

  1. erode the binary WM mask by 1 voxel

maskfilter wm_bin.nii.gz erode -npass 1 wm_bin_ero.nii.gz

  1. substract the eroded image from the binary WM mask

mrcalc wm_bin.nii.gz wm_bin_ero.nii.gz -sub wm_edge.nii.gz


1 Like

Many thanks @NicDC
Your code works very nicely.