[Insight-users] Building shape model to be used with GeodesicActiveCountourShapePriorLevelSetImageFilter

[ramirez at ece.ualberta.ca]

Hi Zach,

Thank you for the reply, especially for the part of using the Danielsson
distance map filter to compute the signed distance from binary images. I
really didn’t think of that one.

Thanks again

Lino

> Based on Leventon's paper that initially described the shape model
> level sets (
> http://www.ai.mit.edu/people/leventon/Research/0006-CVPR/cvpr00.pdf ),
> and looking at the ITK implementation thereof, this is correct.
>
> Note that you need to use the signed level set output -- not binarized
> inside/outside version or single isocontours from the level sets -- as
> input to the ImagePCAShapeModelEstimator. As you probably know, and for
> the benefit of others reading this, the "shape model" training images
> need to be represented as signed distances from the edge of the shape.
> Fortunately, this is the stock in trade of level sets, so you can use
> that output raw.
>
> I was looking around for a class in ITK that could compute signed
> distance maps from a binary image, in case that is all the input one
> has for training images (e.g. input from a non-level-set segmentation
> method). I can't find anything explicitly there (is there??) but of
> course one could create a signed distance image by two applications of
> the Danielsson Distance Map filter to a binarized inside/outside image
> and its complement, and subtracting the two.
>
> Zach Pincus
>
> Department of Biochemistry and Program in Biomedical Informatics
> Stanford University School of Medicine
>
>
> On Aug 8, 2004, at 1:40 PM, Lino Ramirez wrote:
>
>> Dear ITK users,
>>
>> I would like to do segmentation of vertebrae in x-ray images using the
>> GeodesicActiveCountourShapePriorLevelSetImageFilter. I already had a
>> look at the example available in the CVS repository. However, in the
>> example, it didn’t explain how to make the models. For what I have
>> read so far, I guess that the following steps would help me make such
>> a model.
>>
>> 1.- Align the training images
>> 2.- Segment each image using a level-set based method
>> 3.- Use the output level set as the inputs for
>> ImagePCAShapeModelEstimator
>>
>> Do these steps make sense? Or is there some other approach or example
>> I should follow?
>>
>> Thank you and have a nice day
>>
>> Lino
>> _______________________________________________
>> Insight-users mailing list
>> Insight-users at itk.org
>> http://www.itk.org/mailman/listinfo/insight-users
>