[Insight-users] another question on FEM Registration

Luis Ibanez luis.ibanez at kitware.com
Fri, 19 Mar 2004 02:41:43 -0500


Hi Janiv,

The FEM-Based registration method is intended to be used
for computing a deformation field over an image with the
purpose of registering this image against another one.

You could use this method for performing segmentation
if you already had a segmented image and attempt to
register it versus a new input image. This methodology
is also known as 'atlas-based segmentation'.

The mesh that the FEM method creates can have any shape,
however by default, a regular grid is created. No meshing
algorithms are provided in ITK. Meshing is in itself a
hard enough problem.

Once you manage to create your mesh, what the FEM method
does  is to compute forces on the mesh nodes. This forces
are computed from the local displacement that you will have
to apply to a small image region surrounding the node in order
to match the the intensities of the other image.

 From this local forces, and the assumption that the mesh is
modeling a elastic body, the FEM method computes local displacements
in all the mesh nodes.  The displacements are then interpolated
inside the mesh cells.

The nodes in the FEM mesh are not used for segmentation, only
for registration. They original positions are located with
respect to the fixed image coordinate system.

You will find a detailed description of this method in the
Tutorial presentations:

      http://www.itk.org/HTML/Tutorials.htm

in particular at

   http://www.itk.org/CourseWare/Training/NonRigidRegistrationMethods.pdf

you may find useful to also look at

   http://www.itk.org/CourseWare/Training/RegistrationMethodsOverview.pdf


and of course at the SoftwareGuide

    http://www.itk.org/ItkSoftwareGuide.pdf



If you manage to segment your cardiac image, you could
create a Mesh from the segmentation and use it for
representing the elastic body that is going to be deformed
in order to match the other image.

Note that this FEM Mesh doesn't have any real physical
relationship with the cardiac tissue. That is, even if the
Mesh is representing a physical model, this model is totally
unrelated to the physical properties of the anatomic structures
present in your image.



  Regards,



     Luis



---------------------------
b gises wrote:

> hi,
>  
> This is another mail from the previous one on fem registration filter.
> i have 10 time frames of tagged cardiac images, where i have to extract 
> a mesh of the right ventricle. Is it acceptable that i use the 
> itkfemregistration filter for the segmentation of the right ventricle. 
> Does this filter perform registration of images on the user-defined 
> nodes only, or on the whole image. 
> I'm not sure if the user-defined nodes are for the segmentation and 
> extraction of mesh, or for the registration, or for both. i'm wondering 
> if the nodes are placed just on the fixed image, or on both the fixed 
> and moving image. If the nodes have to be place on both images, then i 
> suppose the nodes are used to extract the meshes for both images, and 
> used to register the nodes at time t0 to the nodes at time t1.......Can 
> anyone please explain to me about this filter, and if i can use this to 
> create a smooth meshed model of the right ventricle, and yet get the 
> displacement of the tagged points.
> Thank you very much
>  
> regards
> janiv
>