[Insight-users] registration: time-consuming metric

[Luis Ibanez]

Hi Jonathan,

Yes, the ITK registration framework makes possible to replace
the ImageMetric component.  Note that the other metrics will
not necessarily be faster. Mattes MI use a reduced number of
points, but also has to perform additional computations with
them.

In any case, it is a good idea to experiment with different
metrics, as well as with modifying the parameters of such
metrics, until you find a combination that gives you a good
balance between the computation time and the accuracy of the
final registration.

Such a comparison of Metrics would be an interesting paper
for the Insight Journal. Note that such an experiment should
be reported as a "case study", and should avoid to fall into
generalizations that are not backed up by mathematical proofs.

Please let us know if you have further questions,


   Thanks


     Luis


-----------------------------------
Jonathan Orban de Xivry wrote:
> Hello,
> 
> I'm currently working on deformable registration of RC-CT images in the
> lungs. I'm using a mean square metric with a B-Spline interpolation (on a
> 6 by 6 by 6 grid ) in a 100*100*35 3D image. My computer is a pentium 4
> working with Ubuntu 5.04 and ITK 2.2 version. The optimizer is a LBFGSB
> and I've set the maximum number of iterations to 25.
> 
> The registration works well but takes more than an hour to perform the 20
> to 25 iterations. I guess the metric is time consuming since it is
> performed on the whole image. I've seen that the Mattes Mutual Information
> metric allows to take samples in the image instead of taking all voxels.
> 
> Is there a possibility to do the same thing in other metrics as Mean
> square or cross-correlation metric? Or do I have to implement it by
> myself?
> 
> Thank you for your answer,
> 
> Jonathan
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
>