[Insight-users] metric and optimezer

[Luis Ibanez]

Hi Ines,


The choice of the optimizer depends on how noisy your metric is,
and how big is the misalignment of the two images.


The following are some general guidelines,
but they are not absolute rules by any mean.



a) If you are using Mean Squares, any of the following optimizers
    should be suitable:

     * GradientDescent
     * RegularStepGradientDescent
     * OnePlusOneEvolutionary
     * Amoeba


    Note that this is not an exhaustive list...
    other optimizers may work as well.



b) The Amoeba optimizer is particularly interesting for metrics
    that do not provide Derivatives. This doesn't mean that this
    optmizer only work with these metrics. What it means is that
    the Amoeba optimizer is one of the few optimizers that will
    work with metrics that do not provide Derivatives.



c) I you are using RegularStepGradientDescent as an optimizer,
    any of the following metrics are good combinations:

       * Mean Squares
       * Mattes Mutual Information
       * Normalized Correlation



Please note that what is most important is to fine tune the
parameters of the optimizer for the characteristics of the
particular registration problem.

Keep in mind as well that the "Parameter Scaling" is one of
the most critical aspect of the registration.

What Transform are you using ?

Also, are you using a Transform initalizer ?


   Please let us knonw,


      Thanks


         Luis



--------------------------------------------
imaria.cabezas at telvent.abengoa.com wrote:
> 
> Hi!!I´m Ines and I´m a little confused because the results that I have 
> get them are very imprecise.I work in 3D and with some the same modalite 
> images ( CT) . I need some advices:
> If I use mean squares metric, what is the optimezer  better?
> If I use amoeba, what is the metric better?
> If I use regular step gradient, what is the metric better or more 
> appropiate?
> Thaks you very much.
> Regards!
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users