[Insight-users] Re: General Deformable Registration Question

[Luis Ibanez]

Hi Kevin

Deformable registration method in ITK are not as unified as the
basic registration framework. Most of them have parameters that
you can play with, but don't have interchangeable components
as the basic framework does.

You have two main options:

1) Explore variations of values in the parameters that are
    exposed in the API of the deformable methods.

2) Make fundamental changes in the algorithm implemented in
    ITK. This is probably more interesting but it is also more
    challenging and will require you to become familiar with
    the theory behind the method, and the source code of the
    ITK implementation.


The things that you could change depend a lot on the specific
deformable registration method that you choose.

The main options at this point are:

1) FEM based deformable registration
2) DEMONS deformable registration (a variation of PDEs)
3) BSplines deformable registration (in the registration framework)
4) LevelSet deformable registration


I guess the right way of presenting your question is to
first explore what your actual application is. What are
the constrains of your application, what are the challenges,...
and from that, you could select the methods in ITK that are
better suited to your problem, and evaluate variation that
will make them more appropriate to fit into your constrains.




    Regards,


       Luis


-------------------
Kevin Ming wrote:
> Hi Luis,
> 
> This is a very rudimentary question, so please bare with me:  In ITK, 
> how do I use the existing deformable registration codes to build my own 
> "custom" registration algorithms to suit my particular application?  
> That is, what parameters do I tweak, what components can I change, to 
> make the "general" registration algorithms work for my purposes?
> 
> For example, in the level set deformable registartion algorithm, I can 
> only see 4 parameters to change:  NumberOfHistogramLevels, 
> NumberOfMatchPoints, NumberOfIterations, and 
> GradientSmoothingStandardDeviations.  I could also change the 
> interpolator, optimizer, metric, and transform of the registration 
> framework to add more complexity and customizability, but are there 
> other things I could be doing?
> 
> 
> Thank you very much,
> Kevin