[Insight-users] Registration failure metric

[Luis Ibanez]

Hi Indian Rock,


There are four common ways of validating registration results


1) By the Metric value.
    for each metric you will have to consult the documentation
    in order to identify what is the optimal value. E.g. Mean
    squares is optimal when the value is zero, while normalized
    correlation is optimal when the value is 1, and Mutual Information
    is optimal when the value is equal to the entropy of the moving
    image.

2) By the Transform values,
    This is useful if you created a synthetic example, where you
    misregistered the image on purpose in order to have a "ground
    truth".

3) By Transform stability.
    Analyzing the Jacobian of the Transform in the neighborhood
    of the Transform returned by the registration method.


4) Visualization:
    Combining overlays of the fixed image and the resampled moving
    image. This can be done by combining color channels, or by using
    two displays with paired cursors, or by using a checkerboard
    display.



You will find more details on validation techniques
for image registration in the book:


               Medical Image Registration,
               Joseph Hajnal, D. J. Hawkes, Derek Hill
               The Biomedical Engineering Series.



    Regards,


       Luis


-------------------
indian rock wrote:
> Hi,
> I'm using a translation transform to register two images... Can someone 
> tell me if there's any way to tell if the images registered accurately 
> based on the metric value or some other parameter?
> 
> Thanks,
> IR.
> 
> 
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users