[Insight-users] Re: A question about jacobian modulation.

[Kris Thielemans]

Dear John

> When I was trying to apply Jacobian modulation on nonrigid registered 
> (SymmetricforceDemon) repeat scan, I found that the modulated image 
> (voxel intensity * det(Jacobian)) does not look realistic but the 
> registration result is almost perfect from image subtraction.


yes, I've seen similar things. What seems to happen is the demons etc get lost 
in uniform regions. They essentially work on contours so are happy to put a 
voxel from a uniform region anywhere. I expected that B-splines is a bit more 
well-behaved in that respect, but don't know for sure. People have tried to 
put in penalties to keep jacobian determinants close to 1, but I don't know if 
this is implemented in ITK.

By the way, I'd be interested in seeing how you implemented the jacobian 
weighting in ITK. There's also 
http://thread.gmane.org/gmane.comp.lib.itk.user/13729
to look at.

Kris Thielemans
Hammersmith Imanet, London