[Insight-users] Segmenting Visible Human Data

[Joshua Cates]

Hi Stefan,

We have had good success segmenting localized regions of the visible human
data using the watersheds algorithm (see
http://www . sci . utah . edu/~cates/papers/Cates_MedIA2003 . pdf).  
Unfortunately the ITK watersheds algorithm does not scale very gracefully
to huge datasets.  We did some work on this a while back and our results
can be found in the InsightApplications/StreamingWatershedSegmentation
directory, which streams data through the ITK watershed filters. 

First it is important to understand how the ITK watershed algorithm works.  
We produce an initial (typically oversegmented) transform of regions.  
The initial transform is then used to compute a hierarchy of region merges
that allows us to output segmentations at increasingly coarse levels of
detail.

There are two main factors that limit performance on large datasets. The
first is that the watershed transform is applied globally so huge datasets
produce huge numbers of regions.  This is mainly a memory issue and is
solved by the streaming approach.  The real problem is then in generating
the merge hierarchy.  Computation time increases nonlinearly with size and
we have not found a way to parallellize the merge algorithm.  

Note that computation time and memory consumption of the algorithm is
determined more by the amount of information in a dataset than by the size
of the dataset. (Nonlinear diffusion is an excellent way to reduce detail
and preserve object boundary features.) So whether or not this algorithm
will be appropriate for you depends on the scale of the anatomy you are
trying to segment.

Hope this helps,

Josh.




______________________________
 Josh Cates			
 Scientific Computing and Imaging Institute
 University of Utah
 Email: cates at sci . utah . edu
 Phone: (801) 587-7697
 URL:   http://www . sci . utah . edu/~cates


On Thu, 6 Nov 2003, Stefan Lindenau wrote:

> Hi,
> 
> I have to do a segmentation on a part of the female Dataset of the 
> Visible Human. I think this part will be as big as 4GB.
> After reading documentation and former Mailinglist postings I came to 
> the conclusion that the Watershed Algorithm in conjunction with the 
> SegmentationEditorFltkGUI would be an appropriate approach.
> 
> The goal is to use the segmented data in the VTK to render for example 
> the ribs of the Visible Woman. The SegmentationEditor provides a nice UI 
> to choose the appropriate elements in the segmented image and to 
> manipulate them.
> 
> My concern lies on the memory consumption of the processing. Is it 
> realistic to use this approach with a consumer computer? Maybe another 
> algorithm would fit better to my needs.
> 
> Thank you
> Stefan
> 
> 
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk . org
> http://www . itk . org/mailman/listinfo/insight-users
>