What about the case of non-liner registration of a 2x2x2mm FA map to a
1x1x1mm T1 image of the same subject? In humans, we have done extensive
testing with rigid body and affine images to try to get a good registration,
however the registrations always seem to be a bit off at the genu of the
corpus callosum (though this varies subject to subject) because of EPI
distortion of the FA image in the orbital region. We are most interested in
transforming ROIs from the structural image into native diffusion space to
do FA measurements. We may end up correcting this distortion before
registration, but if this is not possible, will we be able to use non-liner
registration to get a correct registration?
Is this something that FNIRT will be able to do? Also, I heard FNIRT would
be out this autumn, but it's getting pretty cold where I live now. Any news
on when it will be ready? :)
Thanks,
Matt.
-----Original Message-----
From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf
Of Jesper Andersson
Sent: Monday, December 17, 2007 1:51 PM
To: [log in to unmask]
Subject: Re: [FSL] non-linear registration
Hi Cherif,
> Just for education purposes, when would you use 12 dof vs.
> nonlinear (IRTK, FNIRT...) registration?
> I'd be tempted to say that nonlinear reg is best when registering
> across both subjects and acquisition types, or when there are
> inconsistencies between subjects ( e.g. tracts that don't fall in
> the same spot if you "only" use 12dof)... Is that too simplified?
I think it is fair to say that the "true" mapping from one subject A
to another subject B will always be non-linear (if we assume that
there actually is a "true" mapping).
Then the question of appropriate becomes a question of "do we have
good enough information to estimate the non-linear mapping from A to
B?". When we have a high quality scan of A and a high quality scan of
B with the same acquisition type the answer to that question is often
yes, and a non-linear registration will be the better choice.
If on the other hand we have a e.g. high resolution T1 structural
scan of subject A and a low resolution EPI scan of subject B the
answer might be "No, the information we have does not allow us to
estimate non-linear mapping from A to B", and an affine mapping will
be the better choice.
Hope that helps.
Jesper
|