[Insight-users] Mutual Information Question

[Luis Ibanez]

Hi Tao Ju,


Mutual Information is particularly interesting when you are dealing
with multi-modality images. In your particular case it seems that
you have to register two images in the same modality. Even though
MI should be able to register such images, you may get better results
by using some of the same-modality image metrics such as MeanSquares
and NormalizedCorrelation.


About your questions:

1) The reason why FEM uses the MIRegistrationFunction instead
    of the MutualInformationImageToImageMetric is related to the
    fact that FEM only uses the metrics for computing local forces
    applied to the nodes of the mesh.  The MIRegistrationFunction
    actually computes a vector field as output, while the MI image
    to image metric coupes a single floating point value as output.

    The MIRegistrationFunction computes the derivative of the metric
    in a small region around each one of the mesh nodes.


2) You may want to play with the elasticity and density parameters.
    These properties are used for modeling the material associated
    to the image. A lower elasticity will allow for the image to
    have larger deformations under relatively weak forces (comming
    from the derivatives of the image metrics). Relaxing the parameters
    of elasticity and density will allow you to run more iterations.
    Just keep in mind that it is up to you to define when the
    registration is acceptable.


3) A typical approach for registering RGB images is to take on
    of their channels and use any of the registration methods
    designed for scalars. You can extract channels from RGB
    images by using ImageAdpators (as described in chapter 10
    of the software guide), or by using the VectorIndex selector
    image filter.


Regards,


    Luis



-----------------------------
Tao Ju wrote:
> Hi,
> 
> I am new to ITK. I am currently working on registrating two slices of
> nissl-stained brain images. Since the transformation involved would be
> non-global, I tried out the FEM method for deformable registration and I
> wanted to explore what Mutual Information could do. Here are some
> questions:
> 
> 1) First, by looking at the source code, it seems that the mutual
> information image metric is NOT used in FEM deformable registration. In
> FEM registration filter, it MIRegistrationFunction is loaded instead of
> the MutualInformationImageToImageMetric. What's the difference between the
> registration function and the metric? Are there any hints on picking
> parameters for the MIRegistration function? Moreover, when I try to use
> the image metric version of mutual information (by uncommenting the define
> USEIMAGEMETRIC in itkFEMRegistrationFilter.h), the project fails to
> compile.
> 
> 2) Second, I used the FiniteElementRegistrationParameters1.txt
> configuration file, as suggested by Luis. I magnified the
> WidthOfMetricRegion parameter, but the program still jumps out after only
> 3 iterations. Are there other parameters that I need to set to make this
> work?
> 
> 3) What could be the best metric for registering two RGB color images with
> white background?
> 
> Thanks!
> 
> Tao Ju
> 
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk . org
> http://www . itk . org/mailman/listinfo/insight-users
>