[Insight-users] 3D ultrasound to ultrasound registration !!!!

[ilker hacıhaliloğlu]

Hi All

I am working right now on 3D Ultrasound registration of bone surfaces.
First in order to understand the registration framework and also see the
results in 2D I registered 2D Ultrasound slices using
MAttesmutualinformation metric with affine transformation.

I have also despeckled the Ultrasound images before this regiostration. The
algorithm works perfect for my images. On monday I started to work with 3D
images. But the results are really bad.:(

I am using the algorithm given below to scale the parameters.

  parametersScales.Fill( 24.0 );
unsigned int j;
  for ( j = 9; j < 12; j++ )
    {
    parametersScales[j] = 0.001;
    }

  optimizer->SetScales( parametersScales );


I think the parameters need to be tunned again for 3D images. If I change
the scaling values for the first 9 values of parameters from 20 lets say 400
or to a smaller number 0.001 will this mean ( for 400 ) less rotations,
scaling and shearing and scewing and for 0.001 much more rotations scaling
and shearing?????

I have learned from Luis that if I change the parameter scaling for
translation from 1/500 to 1/100000 this will make much more translations and
little rotations. So I think I will not change the scaling parameter for the
translation values.


I will also try the multiresolution registration algorithm and deformable
ergistration algorithms. Could anyone suggest me which one of them would be
much better for registration of
bone surfaces in Ultrasound images? Would it be a better ergistration if I
first use affine transform to get an initial transfrom matrix and then use
the Bsplinetransformation algorithm for my registration problem?

cause in my volumes the surface reflection of the bones are not seen on each
slice. Lets say if I have 150 slices I can only see the surface reflection
of the bones in 90 slices. Would it be much meaningfull to to construct this
slices into a different volume and work with them.???

 Or can I also define a region of interest for mutual information metric in
order to start the algorithm close to an optimal solution . Cause i think It
gets stuck on regions of my ultrasound volumes because of the speckle noise.


Cheers
ilker



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

'.....ich möchte mir eine ader öffnen, die mir die ewige freiheit
schaffte....'

--------------------------------------------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/insight-users/attachments/20060328/ef422ba8/attachment.htm