> does this mean that for a common VBM-Dartel approach the data could be
> imported as affine transformed segmentations and only for shape analysis
> approaches (e.g. classification?) the data should be rigidly aligned?
In principle, but I don't have any plans to incorporate this myself.
Best regards,
-John
> On Mon, 26 Jan 2009 13:59:49 +0100, John Ashburner <[log in to unmask]>
>
> wrote:
> >I wanted to be able to model the shape and size of the brain from the
> >deformations, and not just some measure that has an affine transform
> > factored out from it. In practice though, because DARTEL uses a constant
> > velocity framework, it means that the further it has to deform something,
> > the less accurate are the resulting warps. Beginning with an affine reg
> > may therefore result in more accurate inter-subject alignment, but the
> > shape measures would be a mess to work with. Variable velocity (eg LDDMM)
> > or geodesic shooting methods have fewer such issues - but they are quite
> > a lot slower.
> >
> >
> >My principle aim was not to spatially normalise the brains to MNI or
> > Talairach space. I primarily wanted the brains to be aligned to the
> > other brains in the study, with as little bias as possible. Shape is
> > nonlinear, so if linear approximations are to be used for studying it,
> > then it is more accurate to base those linear approximations at some
> > point close to the average shape. See eg
> >
> >Statistics on diffeomorphisms via tangent space representations
> >NeuroImage, Volume 23, Supplement 1, 2004, Pages S161-S169
> >M. Vaillant, M.I. Miller, L. Younes, A. Trouvé
> >
> >I figured that I would eventually get around to coding up a second step to
> > put all the brains into MNI space and therefore make it easier to use
> > DARTEL for more general spatial normalisation purposes.
> >
> >Best regards,
> >-John
> >
> >On Monday 26 January 2009 10:49, Christian Gaser wrote:
> >> Dear John,
> >>
> >> Dartel is initially using rigidly aligned segmentations. That means that
> >> no size scaling of the segmentations is considered before estimating the
> >> non-linear warps. Although the underlying diffeomorphic registration
> >> method can cope with large deformations, I am wondering why it is not
> >> more appropriate to use affine transformed images (maybe restricted to 9
> >> parameters). The varying scalings of the images might introduce
> >> unnecessary variance/noise. An additional scaling step before warping
> >> might result in much smaller deformations which are needed to register
> >> the brain to the template. Furthermore, if the template is in MNI space,
> >> the resulting images will be to and the postprocessing step of aligning
> >> the images to MNI space could be probably skipped. What is the advantage
> >> of using rigid body transformation rather than affine transformation?
> >>
> >> Best regards,
> >>
> >> Christian
>
> ___________________________________________________________________________
>
> >>_
> >>
> >> Christian Gaser, Ph.D.
> >> Assistant Professor of Computational Neuroscience
> >> Department of Psychiatry
> >> Friedrich-Schiller-University of Jena
> >> Jahnstrasse 3, D-07743 Jena, Germany
> >> Tel: ++49-3641-934752 Fax: ++49-3641-934755
> >> e-mail: [log in to unmask]
> >> http://dbm.neuro.uni-jena.de
|