[Insight-users] About Image Segmentation Strategy Using ITK---asking for suggestions

Luis Ibanez luis.ibanez at kitware.com
Thu Jul 22 10:57:27 EDT 2004


Hi Yu,


1) There are multiple possible strategies for segmenting
   gray matter. From the ones available in ITK you may
   want to try the following

          - Statistical Classification
          - Region Growing
          - LevelSets
          - Watersheds

   as you mention, it is likely that the right strategy
   for your data may involve a combinations of these
   techniques.

   Please be aware that Image Segmentation is *never* perfect.
   The reason is that the information provided by images is not
   sufficient for identifying organs, plus the fact that the
   anatomical defition of organs themselves is often ambigous.

   You cannot get "correct" segmentations, what you can
   expect to get are "segmentation that are good enough
   for a particular purpose". In that regard, you will have
   to define very well what is your purpose with segmenting
   the gray matter from this data set.


2) You should read the Tutorials sessions

         http://www.itk.org/HTML/Tutorials.htm

   in particular the session on segmentation methods

http://www.itk.org/CourseWare/Training/SegmentationMethodsOverview.pdf

   and read the Chapter on Image Segmentation from the ITK
   Software Guide

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



3) Slice-by-slice segmentaton is a bad approach. This is usually
   done as a desperate measure when the segmentation is guided
   by a human operator.  ITK provides N-Dimensional methods for
   image segmentation. You will probably find a combination that
   work properly in a native 3D setup.



4) The human brain is not a very colorful structure, so you probably
   don't get much extra information by using color images.
   However, you can easily double check this by computing the Hue
   component of your RGB data and checking with an anatomist to
   see whether Hue turns out to be specific to the structure that
   you are trying to differentiate.



5) ITK provides LevelSet methods that work on RGB images.

   In fact most of the structures shown on the cover of the
   ITK Sofware Guide book were extracted using the

      VectorThresholdSegmentationLevelSetImageFilter

   That includes the Gray Matter from the Visible Woman
   dataset.

   Note that those filters are not specific for RGB, instead
   they can be applied to any multi-component datasets.



6) You will find useful to look at the validation study
   available in InsightApplications:

          InsightApplications/IBSRValidation

   You will find there methods such as

         - Atlas based segmentation
              - Demons deformable registration
              - FEM deformable registration

         - Statistical Classification.






Regards,


   Luis


----------------
Yu Long wrote:

> Hi all,
> 
>      I have a sequence of human body images which comes from a Chinese 
> Virtual Human Project.( one of them is shown below). I want to segment 
> organs such as gray matter from this image sequence using ITK. But since 
> I have not enough segmentation experience using ITK, I'd like to ask for 
> some suggestions about segmentation methods using ITK. Any suggestion is 
> grateful.
> 
>      In my present plan, I want to use the ITK framework to integrate 
> several segmentation methods within ITK to form a hybrid segmentation 
> method. Is my plan feasible? Are there any recommended methods?
> 
>      The second question.  Shall I do a Slice-by-slice segmentation in 
> 2D space or directly segment in 3D space? I don't know which strategy is 
> better. As I know, most methods in itk support 3d segmentation.
> 
>      The third question.  Shall I segment these images directly in color 
> space or turn them in to gray images first? Some segmentation methods 
> such as levelset methods do not support RGB images, right?
> 
>   
> 
> Thanks your suggestions in advance.
> 
>  
> 
> Regards,
> 
> YU Long
> 
>  
> 
> -------------------------------------------------------
> YU Long
> College of Hydropower and Information Engineering
> HuaZhong Univ. of Sci. & Tech,
> Wuhan China
> 430074
> Tel: +86 27 8754 4644
> 
> 
> ------------------------------------------------------------------------
> 
> _______________________________________________
> 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