[Insight-users] Question about HybridSegmentationFuzzyVoronoi

[Celina Imielinska]

   Jane,

    the most detailed description of the Voronoi diagram classifier
(without the fuzzy connectedness part) you can find in my old paper:

Imielinska, C.; Downes, M; and Yuan, W., "Semi-Automated Color
Segmentation of Anatomical Tissue", Journal of Computerized Medical
Imaging and Graphics, 24(2000), 173-180, April, 2000.

   in the hybrid method that is a combination of (simple) fuzzy 
connectedness and voronoi classification, we use the simplest version of 
otherwise "stand-alone" fuzzy connectedness segmentation (look at other 
fuzzy connectedness filters provided by the itk), to derive statistics for 
a homogeneity operator for the tissue that we are segmenting. We do need 
a well defined homogeneity operator (in theory, it can be provided by
"any" method that can do it "well") to "drive" the subdivisions in the
iterative voronoi classification part of the hybrid method. In the voronoi 
classification, random points are thrown at the image, and each voronoi 
region, in the voronoi diagram, is classified as 
interior/exterior/boundary depending how "close" it is to the 
characteristics of the homogeneity operator. We iterate the method and 
keep subdividing the boundary voronoi regions only, until the method 
converges to the boundary of the object/organ (in the process, we keep 
"pushing" the interior inside-out, and the exterior outside-in, and 
squizze the boundary in-between, until stopping ctriteria "kick-in).

The estimated mean and standard deviation and other parameters that are 
automatically computed from a sample 3D region segmented by the 
(simple) fuzzy, can be stored and applied to a new image (same tissue, 
same image modality etc.). This method hinges on the "quality" of the 
homegeneity operator. We can store the homogeneity operators as a database 
for same tissue/organ, same image modality, etc.

    if you need more details, please let us know (Yinpeng Jin yj76 at columbia 
can answer all questions, too),

  	 -Celina



On Thu, 26 Aug 2004, Jane Meinel wrote:

> Dear itk-users,
> I tried the example of HybridSegmentationFuzzyVoronoi. It is quite good
> image segmentation frame.
> Now I have some questions about this example:
>
> *1. In the example image case BrainT1Slice.png, the parameters are: 140 125
> 140 25 0.2 2.0. Among them, (140, 125) is the seed position. It is 
> obviously. However, "140 and 25 are the estimated mean and standard 
> deviation,  respectively, of the object to be segmented.  Finally, 0.2 
> and 2.0  are the tolerance for the mean and standard deviation, 
> respectively." What do those parameters mean? If I want to segment 
> another image, how should I set those parameters by myself?
>
> *2. In the BrainT1Slice.png case, the voronoi diagram classification
> improves the segmentation a lot after the fuzzy connectedness 
> segmentation step. I want to know details about the voronoi diagram 
> classification. I have read the paper "Hybrid Segmentation of Anotomical 
> Data", which is written by Celina Imielinska, Dimitris Metaxas, Jayaram 
> Udupa, Yinpeng Jin, Ting Chen, and published in MICCAI 2001. But it 
> doesn't describe very clear about voronoi diagram classification. Which 
> paper should I read in order to understand this algorithm better?
>
> *3. I'm impressed deeply by the  figures of the paper mentioned about,
> which show the result of the iterations of VD-based algorithm. How can I 
> draw such pictures by ITK classes? I want to know the procedure in 
> different iterate step of Voronoi Diagram algorithm.
>
>
> Any help is much appreciated! Thanks a lot!
>
>
> Jane
>
>
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Mail - 50x more storage than other providers!