Hi Matt and Sid,
From the probtrackx documentation:
... results from probtrackx can be binned in any available space -e.g.,
diffusion space, structural space or standard space. Note, however, that
tractography itself ALWAYS takes place in diffusion space - it is simply the
results of probtrackx that are stored in the required space.
So unless I am overlooking something, one could FNIRT the diffusion data by the
best possible means to anatomical space and run the probtrackx in the
distorted space with the given FNIRT transformation to yield the results in
the anatomical space. No need to deal with the disputable effects of FNIRT on
bvecs.
What do you think?
Martin
On Tuesday 20 April 2010 02:38:07 you wrote:
> What you are calling spatial normalizations (which I take to mean a
> nonlinear deformation of the image to a template) will introduce
> deformations in the diffusion vector fields. When you are correcting EPI
> distortions, these deformations result in greater fidelity to the true
> anatomy because the bvecs are applied in undistorted space (see for
> example, this poster: http://cds.ismrm.org/ismrm-2004/Files/000087.pdf).
> If you are also applying nonlinear deformations to a template, these
> deformations will not be able to be applied to the bvecs as well and will
> result in
> undesirable deformations in the vector field because the data and bvecs
> will no longer match up. If you applied the nonlinear deformations to the
> template on the calculated values, everything would be fine, the problem
> lies in if you apply them to the raw data and calculate afterwards. If
> you don't care about the fidelity of the vector fields (i.e. are not doing
> tractography), than this concern probably doesn't matter.
>
>
>
> By FNIRT based distortion correction, I mean registering the distorted FA
> to the T1 and then unwarping the diffusion data according to the warp
> field that this generates. This is a last resort method that has various
> problems which have been detailed earlier in this thread.
>
>
>
> Peace,
>
>
>
> Matt.
|