For the SPM5 unified segmentation, it depends on the FOV of your image, but is
in the order of about 1000. There is a prototype of a new version of the
unified segmentation available in SPM8, which uses many more parameters than
that. DARTEL encodes the spatial transforms using about 6,000,000 parameters
per subject.
Note that the actual dimensionality of the deformation is actually rather less
than this. Regularisation causes neighbouring voxels to move together, so
there are really fewer independent parameters. Note also that proper
diffeomorphic registration algorithms (eg LDDMM and geodesic shooting
approaches) actually use fewer still. Strange, but true.
Best regards,
-John
On Tuesday 24 March 2009 19:53, Amir Tahmasebi wrote:
> Dear John and SPMers,
>
> I was wondering what is the dimensionality (number of parameters being
> estimated) of SPM5 unified segmentation-based registration and also DARTEL
> groupwise registration?
>
> thank you,
>
> Amir
|