[Insight-users] watershedImageFilter parameters for 3D CT images

Luis Ibanez luis.ibanez at kitware.com
Sun Nov 1 17:53:04 EST 2009


Hi Gator,

1) Yes, if you are working on a  CT Image, it tends to be useful
     to do some denoising before running a Watershed filter.


2) A rule of thumb is to set the Threshold to be about 1 / 100
     of the Level.


3) Typical values of level are 0.2 ~ 0.3.

     But of course, they vary with the structure that you want to
     extract, and the image that you are providing as input


4)  Please not that usually you want to run the watershed filter
      not in the image directly but in its GradientMagnitude image,
      ... unless... you are lucky enough to be searching for an
     anatomical structure that is dark and it is fully surrounded by
     brighter objects.


5) Finding a set of parameters that will work for a full set of
     images is a challenging problem, and it doesn't have a
     trivial answer.

      What we can suggest you are tools for making this easier:

       We have found effective to use

                a) CMake Macros to define tests for each one
                     of the images in the set, and then run all the
                     tests by using ctest.

                 b) Batchmake (www.batchmake.org).


6)  Note also, that the watershed filter will rarely provide a
      segmentation solution by looking at a  single level.
      It is common to have to construct your segmentation
      by combining fragments that are extracted at different
       water levels.



     Regards,


             Luis


--------------------------------------------------
On Fri, Oct 30, 2009 at 12:38 PM, Gator Philly <gatoratphilly at gmail.com> wrote:
> Hello itk gurus,
>
> In order to effectively use watershedImageFilter, how to optimize those two
> parameters: Level and Threshold?
> If I have a multiple set of 3D CT images, each set of 3D CT images might
> have different voxel intensity characteristics,
> it seems from my experiments that one set of parameters for all sets might
> not the optimal solution. How can I
> choose the parameters based on the characteristics of given set?
>
> The other question I have is should I do some denoising pre-processing such
> as median filter before I do watershed?
>
> Thanks a lot for any suggestions and comments.
> Best,
> Gator
>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Kitware offers ITK Training Courses, for more information visit:
> http://www.kitware.com/products/protraining.html
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>
>


More information about the Insight-users mailing list