[Insight-users] Registration Problems Switch fixed moving images

[Luis Ibanez]

Hi Lars,

Your observation on the asymmetry of the Fixed/Moving image
choice is correct.

When you are registering images A and B, the results when you
choose A as fixed image and B as moving image will be different
from the results that you get when you use B as fixed image and
A as moving image.

The role of the two images is asymmetric during the registration
process.

For example,

   1) The field of view of the fixed image has a dominant
      role over the field of view of the moving image.

   2) The gradients of the fixed image do not participate on the
      computation of the metric derivatives. Only the gradients
      of the moving image are used for computing the metric
      derivatives.

   3) The resolution of the fixed image has a dominant role
      over the resolution of the moving image.


Some of the impact of this asymmetries can be mitigated by fine
tunning parameters of the registration components.

For example, you may be able to get closer to a symmetric behavior
by increasing the number of samples that you use in the estimation
of the image metric, at the price of increasing the computation time
of the registration.


In any case,
              this asymmetry is to be expected, and it highlights
the importance of carefully considering the choice of the fixed image
role and the moving image role. Note that in some applications, there
may be clinical or technical reasons that don't quite leave you a
choice.



    Regards,


       Luis



-------------------
Lars Nygard wrote:
> Hi everybody,
> 
> I have a question concerning registration of a MRT1 image and a MRA (angiography)
> image. First I have registered them using fiducial markers and take this as goldstandard.
> Then I have rotated the MRT1 image around its geometrical center to make some registration
> problems. Im using Versor3DTransform and Versor3DOptimizer and the MattesMutualInformation
> information.
> When I take the MRT image as fixed my results are very good, despite the big rotation and 
> translation i take (various combination of 10 degrees rotation and 10mm translation).
> However when I set the MRA image as fixed and MRT1 image as moving sometimes I get
> wrong results. I was wondering, has this anything to do with how I created the problems or does
> changing fixed and moving images also change convergence characteristics??
> thanks for any suggestions,
> lars
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
>