[Insight-users] FEM - registration with Mutual Information

[Markus Weigert]

Hi together,

I just want to ask briefly, if somebody has
tried the FEM - registration in 3D with Mutual Information metric.
I currently try to register a MR to a CT volume
(thorax, significant missregistration due to lung movement),
but the registration runs only for 2 iterations and calculates
a deformation field with very small values.

The parameters I used were like follows:

rho = E = 1000
gamma = 1.0
timestep = 10
alpha = 1.0
energy reduction factor = 1e-12
Width of metric region = 2
pix per element = 6
number of integration points = 4
Max Iterations = 80

Program gives maxVec = 0.004,
displacement values mostly about 1e-6 !!!

It is a little bit strange, because although the values for E and rho
are very small (with monomodality CT datasets, I got it working 
in a multi - resolution strategy with starting elasticity E = 500.000), 
I'm little bit confused about this behavoir now.
I also tried higher time steps, but this also did not work.
The critical component may be the metric.
So, what is the relationship between the metric and the parameter settings?
Has somebody gained more experience with this filter, perhaps on different datasets
and with different metrics and wants to share his/her experiences with me???

Regards,
Markus








-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20060606/4510b08a/attachment.html