[ITK-users] [ITK] GeodesicActiveContourShapePriorLevelSetImageFilter

[Ken Sutherland]

Hi Everyone,

I have created a shape model of one of the bones of the hand. In the 
training images, some bones are short and fat and some are long and 
thin. Here is an animation of the segmentation process:

https://drive.google.com/file/d/0B30jBBgGrOnrN2otV2t1ZGtpakE/view?usp=sharing

In this example I'm trying to segment one of the long and thin bones. It 
seems to me that the prior shape image should change according to the 
statistics in the three principal component files.

Here is the mean image:

https://drive.google.com/file/d/0B30jBBgGrOnrb09NQ1l2QThvVU0/view?usp=sharing

Thank you for any help.

On 2015/10/23 10:34, Ken Sutherland wrote:
> Hi,
>
> I am trying to use GeodesicActiveContourShapePriorLevelSetImageFilter.
>
> I'm using the knee sample from "Create2DActiveShapeModel", with the 
> mean image and three principle component files. I then use one of the 
> training images as the target, so it should be an easy task. I'm using 
> all the default parameters from the sample code.
>
> I've created an animation by outputting the result image at each step 
> of the registration. The cyan image is the shape model, the white 
> image is the target that I am trying to segment. But it seems that the 
> shape is never changing, only the mean shape image is being used. It 
> shifts left, right, up, down and rotates, but never changes shape, no 
> matter how many iterations I run.
>
> Can anyone provide any insight?
>
> The animated GIF file is here:
> https://drive.google.com/file/d/0B30jBBgGrOnreEZGbDBHcmppZ1U/view?usp=sharing 
>
>
> The log file is here:
> https://drive.google.com/file/d/0B30jBBgGrOnrSjU4RDZvSEJEdm8/view?usp=sharing 
>
>
> Thanks!
>

-- 
Ken Sutherland
Hokkaido University
Graduate School of Medicine
Department of Medical Physics
kensuth at med.hokudai.ac.jp