Hi,
Glad it is getting better.
The help isn't terribly specific I'm afraid, but all the search options
(e.g. nosearch, searchrx, etc) refer to the 8mm search phase only.
They are not constraints on the search, just an estimate of how
wide to try in the initial 8mm search phase. So specifying nosearch
means that only a local optimisation is done at the 8mm phase,
although perturbations (multi-start local optimisation) is still done
at the 4mm phase and progressively increasing dofs are introduced
in the 2mm phase. Each local optimisation can go as far as it needs
to in order to find a local minimum, so this is why it is not a
constraint.
Note that many registration methods only use local optimisation and
do this from a single starting point, which is even more restrictive
than flirt with the -nosearch option.
As for correcting diffusion images, the interaction of motion and
eddy currents effectively leads to matrices which are not purely
8 dof, but can span all 12 dof. For instance, rotation followed by
a scaling is not like scaling followed by a rotation, but end up
looking partially like skews as well (since the relevant "axis" is
tied to the scanner, not the subject). Also note that eddy currents
alone create more than one type of skew. Therefore we use a 12
dof transformation to correct for the combined effect of motion and
eddy currents and I would recommend that you do the same. It is
also why we do not have more constrained schedule files.
Hope this makes things clearer.
All the best,
Mark
On 16 Oct 2008, at 13:38, Franck Lamberton wrote:
> Hi
>
> Thanks a lot, it works definitely better. For example, estimated
> translations and rotation with my old command were huge, about 40mm
> and 80°.
> WIth your options, values are more realistic.
>
> However, I confus with the option -noserach of FLIRT. The help page
> says
> ...
> -nosearch (sets all angular search
> ranges
> to 0 0)
> ...
> and you suggested to use this option to turn off the 8mm search phase.
> Can you precise what is performed with this option ?
>
>
> Now, as I'm working again on DTI, I'm wondering if it's possible to
> impose a
> more correct distortion model with 3 DOF for translation, 3 DOF for
> rotation
> and 1 DOF for skew + 1 DOF for scale only in the Y direction.
> Altogether, it
> makes 8 DOF.
> Does schedule files could be used in this context ?
> If yes, is there any help or comment on these files ? I have been
> looking at
> those included in /usr/local/fsl/etc/flirtsch but it's not quite
> comprehensible.
>
> Thank's again for your help
>
> sincerely
>
> Franck
>
|