[Insight-users] Re: optimization of Active countour

[Luis Ibanez]

Hi Yasser,

This is the reality of segmentation and registration methods:


                   "They have parameters"


That's the reason why any statement about a segmentation process
is useless when it is not accompanied of the parameters used for
running the code.

Variations on the parameters may have dramatic effects on the
output of the segmentation (or registration).

It will be hard to match a manual segmentation by using a level
set. Also, keep in mind that the fact that we call a manual segmentation
a "ground truth" most manual segmentation are relatively poor in shape
and tend to lack coherence from one slice to another. Therefore, the
fact that your segmentation is not perfectly matching a manual
segmentation, is not necessarily a bad thing.


If your images have homogeneous regions in the tumor, then you are
in the lucky situation where a region growing algorithm can be used.
They tend to be faster than level sets. Note that in that situation
you could also use something like the ThresholdLevelSetSegmentation
filter, which gives a good compromise between the simplicity of
region growing and the smoothness control of LevelSets.



    Regards,


       Luis



----------------------
yasser salman wrote:

> Dir sir,
> I am using GeodesicActiveCountor filter to segment the
> Tumor Region in 3D brain volume and calculated the
> number of pixels inside this segmented region, When I
> change the seed point (inside the ROI) the number of
> pixels Changes(5-10%) is it ok?
> , Also the result summed from each 2D slices is not
> equal to result in 3D segmentation and calculation
> (2-8%),  I  am keep  trying to optimize my
> parameters manually to be closely match the manual
> segmentation and calculation  but it takes much time
>  also it depends of the tumor Shapes  so how 
> the operator will know the best and optimum parameters
> values to get the true volume calculations in the
> tumor
> region. at the end is Region growing is accurate more
> than Active countor in tumor volume segmentation
> speacially in homogenous Regions?? Thanks all,
> Yasser salman 
> 
> 
> 
> 
> 	
> 		
> __________________________________ 
> Do you Yahoo!? 
> Yahoo! Mail - You care about security. So do we. 
> http://promotions.yahoo.com/new_mail
> 
>