[Insight-users] Re: Tumor Calculations

Luis Ibanez luis.ibanez at kitware.com
Wed Dec 22 17:50:35 EST 2004



Hi Yasser,


      *ALL* segmentation methods have parameters.

The parameters setting will change the resulting segmentation,
and therefore the volume of the segmented regions.

LevelSets require even more parameters than the region
growing algorithms.

What you have to do is to explore the incidence of these
parameters (for whatever segmentation method you are using)
and find the stable regime where small changes in the parameters
produce minimal changes in the resulting segmentation. Then you
will fix those values.

Note that this must be done over a group of images,
not on a single image.



On your second question, you are addressing the option
of performing atlas-based segmentation, via registration.

In this context, you take a first input image "A", and
perform a segmentation on it, that results in image "AS".
Then you can take a second input image "B", register "A"
to "B" and use the resulting transform in order to resample
image "AS". The resulting image "BS" of this resampling is
a "possible" segmentation for "B".

Note that this approach is only valid when you expect to
find the same sort of structure between the images A and B.
It is questionable whether you should use this for studying
tumors over time.  If you decide to follow this procedure,
you must make sure that you use a deformable registration
technique that allows for compression and expansion, since
you will definitely will want to detect tumor growth or
contraction.

Beware that deformable registration methods also have
*many* parameters to be fine tunned.



     Regards,


        Luis



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

> Hi luis,
> 
> i'm using itkConnectedThresholdImageFilter then
> counting the no of pixels using
> itkRelabelComponentImageFilter it gives me a good 3D
> segmentation result but not accurate when calculating
> the volume coz as u know it's dependent on Seeds and
> Threshold, i switch my work to use
> itkConfidenceConnectedImageFilter but also depending
> on iterations ,radius,...), i need to switch to any 
> Itk segmentation filter less dependent  on these
> factors i'll give a try to Snakes(active contour) but
> i need to know it also it's setting affect on counting
> the pixels like
> itkConnectedThresholdImageFilter,itkRelabelComponentImageFilter,..)
> or it's less dependent and what is the suitable
> settings of the Snake parameters
>  (PropagationScaling,
> NumberOfIterations,MaximumRMSError,...etc( 
> N.B: i'm using ISBR Data as u know .
> Another questions plz. How can I start to segment the
> tumor region using registration in order to calculate
> the volume segmented region? I mean fixed and target
> image in case of my data found or Must have one?
> regard's
> yasser.,
> 
> 
> 
> 		
> __________________________________ 
> Do you Yahoo!? 
> Dress up your holiday email, Hollywood style. Learn more. 
> http://celebrity.mail.yahoo.com
> 
> 






More information about the Insight-users mailing list