[Insight-users] Use of level set methods - setting parameters

Daniele Ranieri ranydany at gmail.com
Fri Nov 25 07:01:27 EST 2011


Dear all,
i have some questions about level set methods as
itkShapeDetectionLevelSetImageFilter,
itkGeodesicActiveContourLevelSetImageFilter,
itkThresholdSegmentationLevelSetImageFilter
and I think i have some problems with the setting of parameters
(propagationScaling, curvatureScaling, advectionScaling , number of
iterations).
I work on CT data dicom of chest (volume 512x512x300 with pixel in
hounsfield unit ) to segment trachea and bronchi using approaches
describing in section of 9.3 in itkSoftwareGuide.
I place 15 seeds inside my object and create map of distance (initial level
set) -> at this point i have my first little problem:

1) Fast marching used to create map of distance takes biblical times if i
don't set stopping value or if i set stopping value at 60,70  (seeds value
for initial distance are -5,-10,-15). Is it normal???
To overcome this obstacle i use stopping value =  5, 10  (luckily trachea
has a diameter of 20-25 pixel).

2) My real problem is when i run level set 's filters : i get resulting
volumes very strange; i get 15 spheres not connected (each sphere grows
where i placed each seed) and these spheres overlap and exceed boundaries
too.
I have tried many combinations of parameters (propagation [-1, 0.5 , 1],
curvature[0.05 , 0.5 , 1 ] and advection [0.1 , 1] )  but no major
improvements!!!
What's wrong? Is it possible have a explanation of how best to use the
parameters of these filters?


3) I get the best result using only the simple itkFastMarchingImageFilter
(as described in itkSoftwareGuide). Is this not absurd? :>
In this case i have my third problem or question : i get my object
segmented inside my real object (trachea) or rather my object segmented
doesn't include the edges.
I have to apply a morphological operator as dilate to obtain approximately
my real object. Is there anything  I can do to improve this aspect???

Thanks for interest and any hints in advance.

Daniele
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20111125/89d2b318/attachment.htm>


More information about the Insight-users mailing list