DTI Brain Mask Including Non-Brain Tissue

Hi. For best tractography, do not register or resample the DWI/DTI. Instead do tractography first in the original data and then you can apply a registration matrix to the tractography if you want it in T1 space.

So for masking, 1) register T1 to baseline. Then 2) apply this transform to the mask that was in T1 space. Then 3) resample the mask to be the same as the baseline.

You don’t ever need to register DWI. baseline image is in DWI coordinate system. So register T1 and mask to baseline and they are registered to DWI as well.

I dont want to register T1 to DWI, but DWI to T1. The T1 is in AC-PC orientation, and I want to keep that for the segmentation.
But for tractography that is not necessary I guess.

Basically I need two “pipelines” then?:
Use the procedure I have for the segmentation (processed with “ugly” mask, resampled, AC-PC orientation), and for tractography: reorient the segmentations to DWI space for use as seeds (good mask, original data)?

sure. just do tractography in DWI space then afterwards you can transform it into any space you prefer. it’s best to avoid interpolation of DWI before tractography if at all possible. No need to modify the data before tractography unless you want to average multiple subjects DWIs or something like that.

you can transform any other data into DWI space for doing tractography. Though the preferred most robust way is to seed in the whole brain mask, then later select the tracts you are interested in. So you could transform the tractography into the space where your segmentations are and use them to select tracts, if that is your goal.

If your goal is to do manual segmentation of the hypothalamus in a standard coordinate system, then yes you would want to register the DWI into that coordinate system and resample. That makes sense for that use case.

Furthermore, if in the paper you will explain this logical reason for resampling the DWI, the reviewers may not complain that you performed tractography after resampling. Just in general this is not something ideal to do.

Thank you very much, that helps a lot :slight_smile:


I found that during “Resample DTI”, the voxel size of the output volume is changed to the voxel size of the “Reference Volume”.
Of course I want to have the same orientation as the reference volume, but not the same voxel size, since my reference volume (T1) has a voxel size of 1x1x1, and the DTI a voxel size of 2.2x2.2x2.3.
If I put no reference volume, the initial orientation is of course no longer approriate and a posterior part of the brain is cut off.
The same happens in “General Registration (BRAINS)”. The output baseline image volume is no longer 2x…, but 1x… Otherwise I would use the registered baseline image as reference volume.
(I followed the steps of the tutorial slides for this procedure).

Why would you want to change the dimensions of the image during registration, and is there a way I can prevent that?

Hi the goal of resampling is always to put an image into the exact coordinate system of another.

If you want a different reference coordinate system you can use a different reference image.

Or, depending on your use case or goal, you can just view the data after registration but without resampling. In Slicer the registration is applied only for visualization without changing the data in this way.

What is your goal here??

Thanks for the answer!:slight_smile:
I now used the option “Manual Output Parameters” in “Resample DTI Volume” with:
Spacing: of DTI
Size: of DTI
Origin: None
Direction Matrix: of T1
For registering the DTI to the T1, I used the procedure from the Wiki. Create Baseline -> General Registration -> Resampling DTI. As far as I understood, there is no way in 3DSlicer to register the DTI to the T1 without “Resample DTI”, even though I dont want to resample. Or what do you mean by “view the data after registration without resampling”?

I think I dont understand the general purpose of changing e.g. the voxel size to the voxel size of another image, and I dont think that it should be done in most cases (for example when coregistering functional to structural MRIs, you dont want to change the voxel size of the functionals). The coordinate system should be correct, yes, in order to get the correct positions, but why the voxel sizes?

Hi that is just the definition of resampling.

You don’t need to resample. Why did you choose to resample?

The images should have been already registered before you resampled, right?

This is what I did:

Resample DTI: We have now computed the registration transform, but the output volume produced above is a registered version of the baseline, which we need for validation only. To get the actual DTI registered we now apply this transform to the tensor image.”
This module is needed for registering DTI to T1. In no other module, I can use the DTI as input for registration.

My overall goal is to coregister/align multiple images with the T1 image, without changing all properties of these images to the properties of the T1, only the alignment. If not necessary, I dont want to resample at all. For example in SPM, you just use the “Coregistration”-Tab for that. You can either reslice or not reslice that (write a new image).

Definitely don’t resample. You should be able to apply the transform to the DWI or DTI in the Transforms module.

Ahh, i did not try that with the DWI… For the DTI, it crashes the program, that’s why I didnt do it.
But with the DWI it works totally fine, thanks!

Please give exact details of how the crash can be reproduced. What exactly did you do and exactly what happened?


More information for you: the masking works correctly in the release version (4.10.2).