Tractorgraphy on MNI space

Hi all,

I am no expert in tractography therefore forgive me if what I am asking is trivial.

I have some DTI HARDI multishell data from a patient with stroke and I would like to understand what’s the best strategy to perform tractography on the tracks affected by the lesion.

This is the situation: we performed the normalization (non rigid registration that takes into account the presence of the lesion) of a T1w MPRAGE scan onto the MNI standard space. We have used this software ( to perform the normalization. The software we have employed also returns the tracks that are likely to be affected by the lesion in the form of a list, and the tracks themselves in MNI standard space. These tracks are probabilistic maps that come from a average population template.

So, what I have is the following:

  1. the T1 scan in standard/MNI space
  2. a probabilistic map of the tracks affected by the lesion (this is in MNI space)
  3. the DTI data

Now, we want to use the probabilistic maps returned by the program as seed for tractography to track WM bundles affected by the lesion. What’s the best approach to do this?

  1. Should I register the DTI data onto the MNI space (I have experience with that so this is not my issue) and perform tractorgraphy from there? Is this valid given b-values/b-vectors are given relative to subject’s space, whereas here I would be working in MNI space?
  2. Should I invert the transformation and register the probabilistic maps onto the subject’s space and perform tractography from there? Even though this is the most lean solution, I am not sure how to use the displacement fields returned by the software (it seems to return only the magnitude of the displacement field in subject’s specific coordinates) and so this is the most un-preferred option at the moment.
  3. Should I first calculate the diffusion tensor in the standard space, and then project that to the MNI space. If this is correct, what’s the function I have to call to perform tractography in MRtrix3 using as input the diffusion tensor as opposed to the multishell DTI data?

Thank you very much, as I said I am a bit new to this and so I am sorry if the question is trivial.


Hi Guilio,

Welcome to the forum!

It sounds like you’re not going to like it, but the best approach is your option 2: project whatever map you want to use for seeding to the subject space of the DTI image. The reason, as you said yourself, is that the b-vectors are defined in native scanner space, and are not trivial to reorient to MNI (or any other) standard space. The toolbox you referred to appears to use ANTs registration under the hood, so maybe you can tweak it a bit to get you the required deformation fields?

Regarding your 3rd option, while there are ways to reorient diffusion tensors (calculated in scanner space) to a template space, we do not have these implemented in MRtrix. We do have a means to reorient FODs through, if you’re willing to take the plunge and use Constrained Spherical Deconvolution instead of DTI.


Hi Daan

thanks a lot. I figured that myself today. The hard way is always the good one :slight_smile: