> probably, I was misunderstood due to a bad formulation of my question. I
> tried to ask: how many additional (transformation-) parameters could you
> include in the SPM-coregistration-algorithm, without losing in accuracy. Is
> the information incorporated in the mutal information (MI)-obtimisation
> function "sufficient enough" to calculate 12 (or even more) parameters
> correctly, instead of the usual 6 parameters, assuming a dataset with high
> resolution, and assuming that, each additional parameter changes the
> optimisation function less or in the same order of magnitude as the
> parameters before.
Sorry about that. I misread your original question.
Again, I think it depends on your images, and the joint intensity histograms
that they generate. The best way of finding out would be to try it. You
would probably need good starting estimates though.
In general, if the voxel sizes in your image headers are correct, then rigid
registration is a more appropriate model to use. An alternative would be to
include some form of regularisation, in order to penalised exessive zooms and
shears, but this is not implemented in SPM (although it is there for the
mutual information registration that can be used to initialise the Segment
function). This would increase the stability of the procedure. The limiting
case of such a model would be an infinite penalty on any zooms and shears.
This would be the same as the rigid-body regostration model.
Best regards,
-John
|