[Insight-users] BSplineDeformableTransform: Initializing fromlandmarks?

[Lydia Ng]

Zach,

> The main advantage appears to be the ability to add a "bulk transform"
> to the B-spline transform, so that you can simultaneously perform
> gross, rigid registration and more fine deformable registration,
> instead of requiring a two-step process. Is this correct?

Actually, the parameters for the "bulk transform" is not optimized. i.e. the
bulk transform parameters are not part of the parameters of the
BSplineTransform class. 

I have been using the "bulk transform" in the following scenario. I first do
a rigid/affine registration, then I use the resulting parameter as the "bulk
transform" instead of resampling. The advantages to me is than I haven't lost
or introduced artificial information and that I can always get a mapping from
the fixed image to the original moving image.

-------------------------------

> Further, I am interested if there is any way to (directly or
> approximately) initialize a B-spline deformable transform with
> landmarks.
-- snip ---
> Any thoughts or musings about either of these questions would be most
> appreciated.

The following is more in the musing category.

I think the KernelTransform hierarchy provides a mechanism to create a
transform from two sets of landmarks (but I haven't tried it myself yet).

So, one thing to try could be to add the KernelTransform as the "bulk
transform".

Alternatively, you could create a field from the KernelTransform by
populating a vector image with (mappedPoint - fixedPoint). Then do some
smoothing or downsampling to make sure you "don't mangle" you data too much.
Then use some sort of adaptor and do a bspline decomposition on each
component of the vector field and then finally convert result to parameters
of the bspline deformable transform.

At this stage I am not sure if ITK has all the components. Let us know what
you find.

Cheers,
Lydia


> -----Original Message-----
> From: Zachary Pincus [mailto:zpincus at stanford.edu]
> Sent: Thursday, June 10, 2004 5:49 PM
> To: ITK
> Subject: [Insight-users] BSplineDeformableTransform: Initializing
> fromlandmarks?
> 
> Hello,
> 
> I am interested in the conditions under which the B-spline deformable
> transform is it more useful than the FEM-based deformable registration
> tools.
> 
> The main advantage appears to be the ability to add a "bulk transform"
> to the B-spline transform, so that you can simultaneously perform
> gross, rigid registration and more fine deformable registration,
> instead of requiring a two-step process. Is this correct? Are there
> other advantages? And what are the reasons one would prefer the other
> deformable registration methods?
> 
> 
> Further, I am interested if there is any way to (directly or
> approximately) initialize a B-spline deformable transform with
> landmarks.
> 
> The only way that I can imagine is to generate a deformation field from
> the landmarks, use that field to deform the moving image, and then use
> the deformed image as the initial moving image for the B-spline
> transform. However, I'm not sure (a) how best to perform the initial
> deformation (1 iteration of a deformable registration filter that
> allows an initial field to be set?), (b) if too much image data would
> get mangled in the process, and (c) if at this point, the B-spline
> transform just isn't the right tool for the job.
> 
> Any thoughts or musings about either of these questions would be most
> appreciated.
> 
> Zach Pincus
> 
> Department of Biochemistry and Program in Biomedical Informatics
> Stanford University School of Medicine
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users