[Insight-users] Image Registration

[Luis Ibanez]

Hi Maureen,

You are working in a very interesting problem.

Since you have a 3DS model, you are correct in considering the use of a
model to image registration method, in which case, the combination of
SpatialObject to Image registration is a convenient option.

You may keep in mind the use of PointSet (from your model) to
Image registration, as an option as well.

The registration process will apply the Transform that you select
to plug into it. Therefore, if you need to perform deformations,
you should consider using the BSplineDeformableTransform.

Please note that in that case you MUST first solve the Rigid
and Affine registration between the model and the Image, and
to use that resulting Affine transform as the "Bulk" transform
to be connected in the BSplineDeformableTransform.


1) Overall Strategy:


                     Start Simple

    do first:

        SpatialObject to Image using a TranslationTransform

    then

        use a Rigid Transform

    then

        use an Affine Transform

    then

        use a BSplineDeformableTransform


2) Nope, unfortunately we don't have similar examples.

    There is a very basic example of ModelToImage registration
    in Insight/Examples/Registration, but unfortunately its
    metric is actually performing the equivalent of an image
    to image registration.

    In your case, you probably want to generate a collection
    of points from your SpatialObject model, and then use
    that collection as the input of a PointSetToImageMetric

    (e.g. to go back to the PointSet to Image scenario).


     The good news is that, since you will be the first one,
     we will be anxiously expecting the submission of your
     paper to the Insight Journal   :-)


3)  You may want to look at the NarrowBand to Image registration
     problem.

All the information about this paper is available in CVS,
on the InsightDocuments/Papers repository:

http://www.itk.org/cgi-bin/viewcvs.cgi/Papers/MedIASpecialIssue/NarrowBandToImageRegistration/?root=InsightDocuments

The PDF version of the paper is at

http://www.itk.org/cgi-bin/viewcvs.cgi/Papers/MedIASpecialIssue/NarrowBandToImageRegistration/NarrowBandToImageRegistration.pdf?rev=1.1&root=InsightDocuments&sortby=date&view=log

and the source code for the tests presented in the paper is
available at:

http://www.itk.org/cgi-bin/viewcvs.cgi/Papers/MedIASpecialIssue/NarrowBandToImageRegistration/Source/?root=InsightDocuments



4) Yes, it does matter what the Fixed and Moving elements are.

    Since ITK visits all the elements of the Fixed object (e.g. all
    the pixels of an image, or all the points of a SpatialObject)
    and maps each one to the physical space of the moving Object,
    then you may in general try to use as Fixed object the one that
    will have less elements.  Typically the SpatialObject.

    ... however... the nature of your application may dictate other
    reasons for preferring the opposite order.



   Regards,


       Luis



----------------------
Maureen Haining wrote:
> Hello All,
> 
> I am looking for some advice with regards to the following:
> 
> I have a 3D geometric "canonical" model of the cerebellum of a mouse in 
> 3DS format. I also have 2D image slice data containing gene expression 
> data and boundary tracings of the cerebellum (embryonic stages). I would 
> like to register these slices to the model. I have read the software 
> development guide and numerous internet postings regarding itk. I am 
> still somewhat unclear about the best strategy for this problem using 
> itk. My initial thoughts are that itkSpatialObjects would support the 3D 
> model and that perhaps applying an affine transform to register the 
> slices to the model would work. Am I correct in thinking that the Model 
> to Image algorithms would deform the model to fit the actual data? 
> 
> 
> 1. Can you recommend a good overall strategy to accomplish this task?
> 2. Is there a similar problem in the Insight Applications I might use as 
> a reference?
> 3. Can you recommend any other good references for this type of problem?
> 4. In itk vernacular does it matter which, the model or the image data, 
> is the fixed image versus the moving image?
> 
> Thanks for any "insight" you can provide!
> Maureen
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users