[Insight-users] Can't get Diffeomorphic Demons Registration to work properly

[Harish M.V.]

Hi everyone,

I'm trying to make the Diffeomorphic Demons class work nicely. (I tried the
few other demons variants - this one works best for my data.)

(1) I've tried it on a bunch of cardiac short axis images - to register the
end systole slices to the topmost slice at end diastole - then I apply the
deformation field Diffeo Demons gives me to a manual segmentation of the end
diastole first slice, to get an automatic segmentation of the end systole
slices. I'm doing all this slice by slice (in 2D).
I have noticed that the registration process does not seem to do well when
the intra-ventricular cardiac muscles are prominent and deform a lot between
slices. This is not good if I am trying to estimate the volume of the heart.
:(
In general, the technique does not do well in capturing details, even if
well defined. (I guess I'm asking for too much from a registration
algorithm.....) Are my observations correct or am I making a mistake
somewhere?

(2) Before Diffeo Demons, I perform a centered rigid transform. However, for
my data, it does not seem to be really needed as the optimum transformation
parameters found are rather negligible. I use it anyway.

(3) To better understand things, I tried it to repeat the experiment for the
binary circular and C-shaped images given here:
http://www.insight-journal.org/browse/publication/154
Unfortunately I have not been able to get the results as advertised even if
I set number of iterations to 10,000. I guess there are two reasons for
this: (a) Something wrong with the images that I recreated (b) I have not
really tried experimenting with the few other parameters available. Maybe
something in those? (c) Some possible mistake in my code. Has anyone tried
these images?

Thanks in advance,
Harish
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20110601/dee647a4/attachment.htm>