Hi Michel,
The problem here really is just that the scalings required are
well outside what is expected in brain imaging, and hence
FLIRT is not optimised to allow this. In fact the schedule of
optimisation specifically assumes that the initial scaling is
relatively close (better than the rotation). You can get quite
a reasonable solution if you use -nosearch (as it stops doing
a large search in rotations *prior* to changing the scaling)
but it still fails to get the exact scaling right (as it seems to
max out around 2.75 whereas it needs to get to about 3.15).
So the main reason for this problem is that it doesn't match
the expectations for what is needed in brain imaging
registration, although you can get pretty close. I got a more
or less exact solution by doing it in two stages (as then the
scaling required in the second stage was very moderate and
only required to soak up the residual error from the first stage).
Hope this helps you out.
All the best,
Mark
On 6 Jan 2012, at 16:46, Michel Thiebaut de Schotten wrote:
> Dear Experts,
>
> Sometimes the easiest things to imagine are the hardest things to realise.
> I'm trying without success to flirt my big box B (B.nii.gz attached) into my small box A (A.nii.gz).
> Any idea why it doesn't work and how to solve the problem?
> Thank you very much for your time and your help.
>
> Best wishes for 2012,
>
> michel
>
> <Boxes.zip>
|