Hi,
I'm not sure exactly what is going wrong here, but some things
you can try are:
- Put a "-out" option in the first flirt call and see what that image
looks like. It should be well aligned - otherwise the second
call is pointless.
- If it is the first flirt call that is failing (which would be my
guess)
and the images are initially reasonably close to a good
alignment (not more than 20 degrees or 10% of the
FOV in translation out) then try adding a "-nosearch" option
to the first flirt call and see if that helps.
- Check that your original images are OK in FSLView - particularly
the voxel dimensions (make sure it doesn't look squashed
or expanded in any direction).
- Try "normmi" or "corration" as a cost function instead of
"mutualinfo" and see if that helps.
- Try switching the "ref" and "in" volumes to see if the registration
works better the other way around. If so you'll then need to
add a middle stage that inverts the transformation before
applying it in the last call (where you would keep the ref and
in volumes the same way around as you are currently doing).
The inverting call is:
convert_xfm -omat xx_inv.mat -inverse xx.mat
- If you've got any features that show up in one image and not
the other (e.g. tumors/lesions) then try using a weighting
volume on either the input or reference space (see the
FSL course lectures and practicals for more information).
Hopefully one of these things will help.
I would try them in the order I've written them above.
All the best,
Mark
On 6 Feb 2009, at 18:53, Mi Jung Kim wrote:
> Hi,
>
> I am using FLIRT linear registration tool for generating T2map.
> I've got into several problems while I have been doing it. Besides
> complete
> registration failure, I have seen 'blank', somehow 'mirror images in
> each
> slice',or 'no matching with reference images'. One of examples is
> attached.
>
> Here is what I did for my analysis.
>
> flirt -ref xx -in xx -out xx -dof 6 -cost mutualinfo -omat xx.mat
> flirt -ref xx -in xxx -out xxxx -init xx.mat -applyxmf
>
> Is anybody who has seen something like this problem?
>
> Thank you for your time,
>
> Mi Jung
>
> <Picture 1.png>
|