On Wednesday, August 31, 2011 02:16:31 you wrote:
> On 31 Aug 2011, at 02:49, Benjamin Kay wrote:
> > Sorry for the newbie question, but how can I limit the search angle in
> > flirt? I am trying to do a 7 dof registration of a low-resolution
> > functional volume to a high-resolution anatomical volume. Nominally
> > there should be less than a degree of rotation about the
> > anterior-posterior axis, but flirt is adding approximately 5 degrees of
> > rotation. I had thought I might fix this with "-searchry -1 1" or
> > "-searchry 0 0", but neither of these seems to have any effect.
> Dear Benjamin,
>
> The search bounds only limit the range of the initial starting positions
> for the search. They do not constrain the final transformation, and so
> you cannot use them to do this. We do not provide a constraint option
> as if the cost function is telling the registration method to try larger
> and larger angles and you put a constraint on this then you will simply
> get back your constrained value, which you have chosen arbitrarily and
> hence is not a good solution. It is far better to determine why there
> might be a problem in the registration and fix it that way.
>
> Have you got fieldmaps?
> Have you tried other cost functions?
> Do you really need 7 dof or will 6 dof be enough?
> Is there much bias field in your anatomical image?
> Have you run brain extraction?
>
> These are the kinds of things I'd explore first.
> All the best,
> Mark
>
Dear Mark,
Thank you for explaining that. It sounds like the -searchr* options are meant
for performance optimization and not as search constraints. Looks like I need
to use nudge.
As for the checklist, no, unfortunately I do not have fieldmaps. Yes I had
tried different cost functions: corratio, mutualinfo, and normmi. No I do not
necessarily need 7 dof, but the registration goes equally poorly with 6. Yes,
I had run brain extraction on the anatomical image. Brain extraction on the
functional image has no effect. No, there is not more artifact than usual in
the anatomical. There is a small subarachnoid cyst, but I had exluded it with
a binary mask to -refweight.
Benjamin
|