[Insight-users] How to use multiple seeds in level set segmentations?

Luis Ibanez luis.ibanez at kitware.com
Mon Aug 9 11:19:02 EDT 2004


Hi Weiguang,

Most of the LevelSet methods in ITK do not expect seed points
as inputs, they actually expect an initial LevelSet.

The only LevelSet method that expects seed points as inputs is
FastMarching, and it accepts any number of them. Please look at
the doxygen documentation for this class

http://www.itk.org/Insight/Doxygen/html/classitk_1_1FastMarchingImageFilter.html

In order to use multiple seed points, just add more points to
the NodeContainer that you pass to the FastMarching filter.

It is common to use FastMarching as a mechanism for initializaing
a more advanced LevelSet method. In that context you provide seed
points to FastMarching, execute it, get a rough level set as output
and use it as initial level set for a method such as ShapeDetection
or GeodesicActiveContours.

Please read the Chapter on LevelSet segmentation from the ITK Software
Guide. These concepts are explained there in detail.

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


BTW, You are right, bone is not trivial to segment. As with any other
anatomical structure, you get 95% segmented correctly in the first few
minutes, and the remaining 5% may take several months of fine tunning.
One of the issues with bone, (as with other anatomical structures) is
that the definition of what "bone" is, happens to be quite ambiguous.
You need to specify if you are looking for cortical bone, travecular,
bone, marrow cavities, how much you care about cartilage and how much
you care about the epiphyses where calcification is lower. As with
Image Registration, Image Segmentation can only be succesful when you
have an application at hand that tells you 'how much' is 'good enough'.


Regards,


    Luis


---------------------------
wyao4 at uwo.ca wrote:

> Hi,
> 
> The procedures that ITK provides such as GeodesicActiveContourImageFilter seem
> accept one seed only. This could make the performance of the filters quite poor
> in segmenting objects like bone(I am not kidding, bone is hard to be segmented
> because of its inner structure). However, if we give more seeds the performance
> could be better. Do any body know how to give multiple seeds for these filters?
> Thanks.
> 
> Weiguang
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 






More information about the Insight-users mailing list