FBA in MNI Space



Hi MRtrix experts,

Is it possible to use fixel-based analysis and report the results in MNI space? Presumably transforming DWI data, or FOD data, is ill-advised as this will affect the orientation of the data. The reason I wish to use MNI space in this case is to report the location of significant regions within a known co-ordinate space.




Hi Matt,

We’re working on better mechanisms to do this sort of thing, but for now you can definitely get it done with existing tools.

Since you’re only interested in the 3D spatial location of effects, and not orientation, you don’t actually need to worry about preserving orientation information for this particular purpose. I would just reduce the fixel-wise thresholded result in your study-specific template space to a voxel mask using the fixel2voxel command, and transform the resulting binary mask to MNI space (based on having registered your population template to the MNI template). If you have multiple significant pathways, you could break them up using maskfilter connect, followed by mrcalc -eq to get a 3D binary mask corresponding to each connected component.

I do have an MRtrix3 command sitting here for calculating the centre-of-mass of an image, which would take such 3D binary masks and give you a single 3D location for each. Trouble is, I can’t decide what to call it :sweat_smile:



@rsmith, I am sorry for posting here. Based on this comment: "….transform the resulting binary mask to MNI space (based on having registered your population template to the MNI template)…"
I have a population template of 42 patients and I used the masks option. I was wondering how to register this population template with the standard space to be able to use several atlases that are in MNI space. This would be useful for trace different pathways of interest.
When I tried to register my population template to the MNI space I had the following error: input images do not have the same number of dimensions. And in visual inspection the population template seems to be smaller than any MNI standard T1.
Any advice would be greatly appreciated

Thank you


Hi Josue,

I think there’s a few concepts being either conflated or misunderstood here:

  • Using the masks option when building the population template is not really relevant to the discussion here. That allows you to provide a brain mask per subject for the template building process. Here we’re specifically interested in transforming masks between two different template spaces.

  • If you have tried to register your population template to MNI space, there are two issues here:

    1. Your FOD template is stored as spherical harmonic coefficients, whereas the MNI template is only a 3D image; the command is therefore unable to determine unambiguously what information it is supposed to use to drive the registration.

    2. The mrregister command currently only possesses a sum-of-squared-errors similarity metric, which is inappropriate for registering images with vastly different contrasts (which would be the case when comparing the FOD l=0 image to the MNI T1-weighted image).

    What you would need to do here is:

    1. Explicitly extract the l=0 image from the FOD template, using mrconvert -coord 3 0.

    2. Use some other software package that is capable of using more sophisticated registration similarity metrics to perform the registration between the two template images. Results from such registration can then be converted into MRtrix3-compatible format in order to perform image transformations (whether warping results from your population template to MNI space, or warping parcellations defined in MNI space into your population template space).



Thanks Rob