[Neuroimaging] [Dipy] Should we resample dMRI to T1w, or T1w to dMRI?

[Eleftherios Garyfallidis]

Hi Robert,

Great question. What I personally prefer doing is first reslice the DWIs to
1mm^3 and then bring the T1 to diffusion resliced 1mm^3 space . In that way
your T1 is not going to be damaged from low voxel resolution. Also you
don't need to re-orient your b-vectors as no rotations were introduced.

Of course there is some equivalence between going to a resampled dMRI and
going to T1. This is a question I think more about which modality will be
more capable in the future. I think that DWIs because they are 4D data even
if their initial voxel resolution maybe lower than the T1 after being
"smartly" denoised they can reach the resolution of T1 or even surpass it.
Also it has been recently shown that you can create T1 looking images
directly from dMRI datasets. So, it is very likely that gathering a T1 in
the future may not be as important as it is today.

Also, I further think that registration of dMRI datasets can use more
information to correctly register different subjects because you can
additionally use local orientations etc. Nonetheless, whatever you decide
it is not the end of the world you can always go forth and back from the
two spaces. But make sure your DWIs are resliced to 1mm^3 this helps both
with registration but also with tracking.

I hope this was helpful.

Best regards,
Eleftherios









On Tue, Nov 24, 2015 at 9:38 AM, Reid, Robert I. (Rob) <Reid.Robert at mayo.edu
> wrote:

> Hi,
>
> My apologies if you've already seen this on the FSL mailing list - I'm
> trying to get a wide sample.
>
> We are trying to decide on whether our lab's internal diffusion MRI
> processing should write its images in dMRI space or T1w space, and would
> appreciate your opinions.  The question is prompted by the EPI undistortion
> step, which necessarily introduces a resampling step of some sort, and
> tends to produce results in either T1w space (e.g. BrainSuite) or in
> DTI-sized voxels (e.g. FSL's topup).  Which algorithm we use depends on
> what other data we have (topup works best when pairs of images with flipped
> phase encodings are available, but usually they are not), but we would like
> to settle on a consistent format, and avoid resampling more times than
> necessary for our typical analyses.
>
> Theoretically each additional resampling step introduces some degradation,
> so "native" space is the right one for dMRI.
> However,
> * The T1w voxels are so much smaller than the diffusion voxels that the
> resampling of dMRI into T1w space is typically excellent.  If a diffusion
> processing step is using T1w (for example, tractography often uses the T1w
> gray/white boundary to place seeds), it seems like the T1w image would be
> more damaged by going to DTI than the DTI would be damaged by going to T1w.
> * The undistorted diffusion images *are* resampled, so they are not really
> in native space.   If resampling is necessary it is better to resample
> finely (i.e. use T1w space).   On the other hand, many parts of the brain,
> like the parietal lobe and motor-sensory strip, are relatively unaffected
> by EPI distortion and may be left in nearly native space after undistortion
> with topup, if head motion and eddy current distortion are not a large
> problem.
>
> Based on feedback I've already received, I should add that we will be
> doing quality control on the T1w <-> dMRI registration, and would fix,
> reject, or if necessary work around drastic registration failures.
>
> We look forward to hearing from you,
>
> Rob
>
> --
> Robert I. Reid, Ph.D. | Sr. Analyst/Programmer, Information Technology
> Aging and Dementia Imaging Research | Opus Center for Advanced Imaging
> Research
> Mayo Clinic | 200 First Street SW | Rochester, MN 55905 | mayoclinic.org
>
>
> _______________________________________________
> Neuroimaging mailing list
> Neuroimaging at python.org
> https://mail.python.org/mailman/listinfo/neuroimaging
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/neuroimaging/attachments/20151124/3c38f8c5/attachment.html>