Registration FLAIR -> dMRI vs. dMRI -> FLAIR, different resolutions

Hi everyone,

I’ve asked this on the FSL mailing list already but just got a recommendation and not a more detailed explanation so I was wondering if anyone here might have thoughts on this:
I’m trying to bring FLAIR, T1w and dMRI images into the same space in order to use WMH segmentations (created with TRUENET) on maps of dMRI metrics which are mainly free-water corrected DTI metrics. I can see a few options and while they all work, I’m wondering if there are theoretical reasons why one might be better than the other, especially with the different resolutions in mind. The FLAIR and T1w images have an isotropic resolution of 1mm, the dMRI images a resolution of 1.75mm.
The options I see are

  1. move the FLAIR and T1w images to dMRI space (registering them to B0 images) and run TRUENET on the transformed images in the lower resolution, or
  2. run TRUENET on the FLAIR images in original resolution and transform the resulting WMH masks to dMRI space via the FLAIR images, or
  3. move B0 images into FLAIR space and run TRUENET in the original higher resolution. If option 3 is chosen, would it be better to a) transform maps to FLAIR space by using the transformation matrix of the B0 images on them, or b) generate the maps in FLAIR space after transforming the dMRI images?
    I preferred option 1 but I’m wondering if in theory downsampling or upsampling is better, and if it’s the mask or the target dMRI image that should be preserved and not resliced? Of course there’s also the practical reason that the dMRI images have over 100 volumes each, which makes option 3b unattractive.

Just wanted to make sure that this is correct - thank you in advance!
Tanja

In general it’s best to transform masks because you are just using nearest neighbor interpolation. Less concern about interpolating your brain images.