Dear John,
I tried to make it work, but I still get no useful results. A mouse brain is
about 20mm long (anterior-posterior), so I would use a kernel of 0.8mm.
Here are my defaults so far:
% Spatial Normalisation defaults
%=======================================================================
defaults.normalise.estimate.smosrc = 0.8;
defaults.normalise.estimate.smoref = 0.8;
defaults.normalise.estimate.regtype = 'subj';
defaults.normalise.estimate.weight = '';
defaults.normalise.estimate.cutoff = 25;
defaults.normalise.estimate.nits = 16;
defaults.normalise.estimate.reg = 1;
defaults.normalise.estimate.wtsrc = 0;
defaults.normalise.write.preserve = 0;
defaults.normalise.write.bb = [[-128 -128 -163];[127 127 162]];
defaults.normalise.write.vox = [0.5 0.5 0.5];
defaults.normalise.write.interp = 1;
defaults.normalise.write.wrap = [0 0 0];
The normalisation template currently has a voxel size of 50 microns cubed,
with a matrix size of 256x256x326 (has not been smoothed yet).
One of my source images has a voxel size of 40 microns cubed, and a matrix
size of 256x256x512.
When I try now to do the spatial normalisation, I get a matrix size of
511x511x651, and the brain itself is tiny. Matlab is giving out a message
that the FOV is too small for nonlinear registration.
In the long run, I would like to use the real voxel sizes, rather than
manipulating them. In addition, I would like to take advantage of the
nonlinear part as well. Do you see a possible fix for doing that?
Thanks very much for your help in advance. Looking forward to hearing from you.
Regards,
Torsten
|