| 1. When it does the realignement, which is the order of the operations:
| first translations, second rotations?Or the inverse?
The realignment estimates all six parameters at the same time. These
parameters are specified as described in spm_matrix.m, and the actual
voxel-to-voxel mapping is given by:
M = inv(M2)*spm_matrix(p)*M1;
where:
M2 is the matrix given by spm_get_space of the moved image.
M1 is the matrix given by spm_get_space of the reference image.
At the start, M2 is typically the same as M1
p are the six parameters.
A voxel [x1 y1 z1] in the reference image is mapped to a voxel
M(1:3,:)*[x1 y1 z1 1]' in the moved image.
After realignment, M2 is updated for the moved image such that
M2 := inv(spm_matrix(p))*M2;
The parameters saved after the realignment are derived by:
q = spm_imatrix(M2/M1) % After M2 has been updated as above.
p can be derived from: p = spm_imatrix(M1/M2)
|
| 2. Where does the coordinates frame has the origin?
The origin of the coordinate frame of the reference image is at voxel:
inv(M1)*[0 0 0 1]'
The first voxel in the images is considered to be [1 1 1], rather
than [0 0 0] (or even [0.5 0.5 0.5]).
|
|
| 3. Do you think that's possible to use "Coregistration" for heart
| images? I mean, the segmentation method is it dependent of the
| anatomical features?
In SPM99, hidden under the <Defaults> button is an option to use
mutual information coregistration, which may work for cardiac images,
providing there is not too much nonlinear distortion.
Best regards,
-John
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|