[Insight-users] ImageRegistration4 example problems

[Luis Ibanez]

Hi Fernando,

The ImageRegistration4 example is illustrating the use
of MattesMutualInformation metric. This metric is well
suited for multi-modalilty registration.

Mutual Information, on the other hand, will not behave
well in same-modality registration. Altough in theory
it could do it, in practice the joint-probability
distribution of gray levels is concentrated into a small
region of the joint histogram (the diagonal...) making
that the estimations mutual information to be poorly
evaluated.

----

Please use this example with images of different modalities.

For example,
you may want to register the following two images :

    -  BrainT1SliceBorder20.png
    -  BrainProtonDensitySliceShifted13x17y.png

using the following command :

   ImageRegistration4.exe BrainT1SliceBorder20.png
       BrainProtonDensitySliceShifted13x17y.png   NonJunk.png


(in a single line...)

will produce the following output:


> 0   -0.513644   [1.63518, 3.65051]
> 1   -0.583664   [-0.68964, 6.90554]
> 2   -0.537496   [-0.331539, 10.8895]
> 3   -0.55033   [1.5392, 10.1821]
> 4   -0.584612   [3.43558, 10.8176]
> 5   -0.61587   [3.97688, 12.7429]
> 6   -0.635517   [5.94851, 13.0786]
> 7   -0.667189   [5.97992, 15.0783]
> 8   -0.698122   [7.8247, 15.8509]
> 9   -0.769544   [9.42667, 14.6535]
> 10   -0.795233   [11.4087, 14.921]
> 11   -0.944684   [12.8385, 16.3195]
> 12   -1.08362   [12.9449, 18.3167]
> 13   -1.0092   [13.1342, 17.3348]
> 14   -1.13406   [12.1944, 16.9931]
> 15   -1.03496   [12.6933, 16.9606]
> 16   -1.1376   [13.1932, 16.9521]
> 17   -1.15311   [12.9794, 16.8226]
> 18   -1.13864   [12.9714, 16.9474]
> 19   -1.13672   [13.0945, 16.9691]
> 20   -1.14475   [13.0534, 17.0163]
> 21   -1.1445   [13.044, 16.9865]
> 22   -1.13925   [13.0564, 16.996]
> Result =
>  Translation X = 13.0564
>  Translation Y = 16.996
>  Iterations    = 24
>  Metric value  = -1.13974


The Mattes implementation of MutualInformation
is much less noiser than the Viola-Wells
implementation.


Both metrics where discussed in detail by
Lydia Ng, in her chapter on  registration
in the SoftwareGuide.pdf

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

Section 7.4, pages 182 to 189.

----

Please let us know if you have further questions,


   Thaks


     Luis



-------------------------------------
Fernando Rannou wrote:
> Hello Luis
> 
> Something is wrong with ImageRegistration4 example (Insight 1.2).
> Here is the output:
> 
> $ ImageRegistration4 BrainT1SliceBorder20.png 
> BrainProtonDensitySliceShifted13x17y.png  junk.png
> 0   -0.497728   [-3.98581, -0.336661, 0]
> 1   -0.442616   [-7.91193, 0.42857, 0]
> 2   -0.421535   [-5.97582, -0.0729125, 0]
> 3   -0.44281   [-4.14183, -0.870721, 0]
> 4   -0.443689   [-4.88292, -2.72835, 0]
> 5   -0.413782   [-5.58482, -2.01609, 0]
> 6   -0.423873   [-6.05154, -1.13168, 0]
> 7   -0.431457   [-6.43308, -0.207324, 0]
> 8   -0.435543   [-7.37642, -0.539157, 0]
> 9   -0.427424   [-7.25227, -0.0548157, 0]
> 10   -0.423302   [-7.7102, 0.145922, 0]
> 11   -0.426716   [-7.46115, 0.167688, 0]
> 12   -0.428633   [-7.26913, 0.00760221, 0]
> 13   -0.424174   [-7.39149, 0.033136, 0]
> 14   -0.426957   [-7.32955, 0.0414921, 0]
> 15   -0.429356   [-7.34259, 0.0698952, 0]
> 16   -0.430093   [-7.327, 0.0687898, 0]
> Result =
> Translation X = -7.327
> Translation Y = 0.0687898
> Iterations    = 18
> Metric value  = -0.430028
> Aborted
> 
> Notice that the Translations are way off, but for some reason it ran 
> only 16 iterations.
> Also, after printing the Result, it aborts.
> 
> How can I set the convergence criterion?
> 
> Thanks
> Fernando
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users@public.kitware.com
> http://public.kitware.com/mailman/listinfo/insight-users
>