Matthew Brett wrote:
> I have just started using itk (www.itk.org) with SPM images, to try
> out their registration algorithms. Has anyone got any experience
> with this?
We're partway through the process of integrating ITK into our VoxBo
package, and we've generated a handful of sample results just to see
if things work. So far, we've experimented with Demons-based
inter-subject coregistration and rigid (intra-subject) coregistration
with various metrics (including mutual information). We haven't
evaluated them formally yet, but to the naked eye, the results seem
encouraging (e.g., averages of Demons-registered high-res anatomical
volumes are reasonably sharp, rigid inter-modality registration
nestles nicely). The code is fairly slow, but the version we're using
at the moment (not the latest) seems pretty stable.
This is a project we're doing in collaboration with Jim Gee's lab here
at Penn -- they're directly involved in the ITK project, but my own
knowledge of the ITK code is fairly superficial (so most of the "we"
above is actually a guy in Jim's lab). When our code is ready for
general use (I hope before this fall, some executables are close), it
will work transparently with analyze, nifti, and a few other file
formats, so it should be easy to integrate with an SPM pipeline. Our
applications are just executables you can run from the command line on
Linux and OSX. We'll probably release the applications before we have
any kind of formal evaluations. But if there's interest, I can make
some test results available in the meantime.
dan
|