Which warp to use for registering FA maps?

Hi all,

I’d like to see whether using information from FOD registration is beneficial for registering FA maps into a common space.
I have two sets of warps as follows:
1- warps in -warp_dir created by population template when creating a wmfod_template.
2- warps created by mrregister with -nl_warp switch when registering each individual wmfod into the wmfod_template created in step1.

my question is which one these warps are more suitable for registering FA maps into a common space?


I would be thankful if anyone could share their thoughts about this question.

Thank you!


As far as I understand, you should use the transformation warp “subject2template_warp.mif” produced from registering each individual subject to the template. You basically need to repeat step 13 from the “Fibre density and cross-section - Multi-tissue CSD” documentation. So:

foreach * : mrtransform IN/wmfod_norm.mif -warp IN/subject2template_warp.mif -noreorientation IN/fod_in_template_space_NOT_REORIENTED.mif

But replace fod_in_template_space_NOT_REORIENTED.mif part with your FA map. And make sure you keep the -noreorientation flag.

I hope that helped (and I hope it’s actually correct :sweat_smile:)


Thanks, Joe.
What is the reason for using the -noreorientation option?
I actually used:

mrregister IN/wmfod_norm.mif -mask1 IN/dwi_mask_upsampled.mif ./template/wmfod_template.mif -nl_warp IN/subject2template_warp.mif IN/template2subject_warp.mif

mrtransform IN/FA.mif -warp IN/subject2template_warp.mif IN/FA.wmfod_templatespace.mif

We distinguish between registration, the process of determining the transformation that aligns images, and transformation, the processes of transforming (and regridding if required or specified) one image into another space using a given fixed linear or nonlinear transformation.

Registration parameters and the algorithm sequence (initialisation) differ, hence you will end up with different transformations from the group-wise iterations in (population_template) and direct pairwise registration of subject and template data (mrregister). If you want to use images that were not used in the template creation, you’ll need to use the latter approach. Otherwise, you can use either warps but I do not know which one is better suited.

The -noreorientation option currently has no effect if the number of volumes does not match that of a spherical harmonics image but that will change in the next version. In the FBA pipeline, we recommend to transform images without reorientation, fixelate, and then reorient the fixels to avoid FOD segmentation artefacts in the fixel calculation step. For template creation, you’d typically want the FODs to be reoriented when warping FODs to preserve directions and connectivity between adjacent voxels but this is not necessary for scalar quantities such as FA.

1 Like