For aligning to some form of standard atlas, I'd suggest that you use
some form of spatial normalisation, rather than the coregistration.
Brains of different shapes and sizes don't register very well using a
rigid-body transform.
In principle, you are talking about doing some form of model
selection. This is something that tends to have been largely ignored
for my image registration routines. The only work that I can think of
which does something similar is:
Simpson, Ivor JA, Julia A. Schnabel, Adrian R. Groves, Jesper LR
Andersson, and Mark W. Woolrich. "Probabilistic inference of
regularisation in non-rigid registration." NeuroImage 59, no. 3
(2012): 2438-2451.
However, this work does not actually produce a measure of model
evidence because of the technical difficulties of computing the log of
the determinant of the rather large matrices involved.
Best regards,
-John
On 29 October 2012 11:59, Fred Sampedro <[log in to unmask]> wrote:
> Dearest SPM experts,
>
> I'm fitting a medical volume to a couple of hand-made atlas using the SPM Co-register tool. I would like to check whether the medical volume fits better with the first or the second atlas, i.e (pseucode)
>
> performance_1 = coregister (volume, atlas_1);
> performance_2 = coregister (volume, atlas_2);
>
> if (performance_1 > performance_2): use atlas_1 ; else: use atlas_2
>
> Any suggestions? Is there an output variable of the co-register process that could be used for this task?
>
> Thanks a lot in advance!!
>
> Fred Sampedro, Faculty of Medicine, University of Barcelona
|