[Insight-users] Use of advection image to guide geodesic level set

[Amardeep Singh]

Dear ITK Users

I think, I have found the reason of my problem and I would like to post 
my solution, in case others encounter the same situation.

After digging through the code, it seems to me that the 
SetAdvectionImage() method does not have any effect on the 
GeodesicActiveContourLevelSetImageFilter. The advection image is always 
calculated from the feature image in the CalculateAdvectionImage() 
method of the filter,  independently of having set an advection image 
before.
Is this supposed to be the desired behaviour in all cases? I hacked the 
code and now use a flag in order to ensure that my own
advection image is used, in case the SetAdvectionImage() function has 
been called. Everything works fine now and the level set locks to the 
inside of the ring.
In case the SetAdvectionImage() is supposed to have an effect: Shall I 
file a bug report?
Perhaps someone can comment on this.

Best regards
Amar


Amardeep Singh wrote:
> Dear ITK Users,
>
> first of all, thank you very much, Luis, for answering my questions on 
> the gradient vector flow (GVF). This
> really helped to lift some confusion. Unfortunately, I am stuck again 
> with my experiments on (at the moment) synthetic images and so I would 
> like to ask for the help of the mailing list.
> My scenario is the following:
> I am dealing with 2D images and consider a bright ring on a dark 
> background. I use the ring as  an edge map and
> calculate the gradient vector flow from it. I start with a circle that 
> is inside of the ring and want to propagate it to the center of the 
> ring by using the gradient vector flow as an advection term by using 
> SetAdvectionImage.
> My speed image is 1 inside of the ring, 0 in the area of the ring and 
> -1 outside so that the level set is simply guided by the advection 
> term on the ring.
> My problem is that the level set locks to the outer border of the 
> ring, not to its center, which is really confusing me.
> I know that deformable models are very sensitive to the parameter 
> settings, but in my case I would think that this sensitivity should be 
> a little less important as the speed image is 0 on the ring and only 
> the advection term is important. I checked that the 0 isoline of the 
> GVF is where I want it to be, in the middle of the ring.
> All images have direction cosines of a unit matrix.
> My parameters:
> GVF:
> noise level: 20000 (in order to get a very small time step)
> iterations: 500
> LevelSet:
> iterations: 1000
> RMS error: 0.01
> propagation scaling: 1
> curvature scaling: 0
> advection scaling: 1
>
> I would be very happy if anyone could shed light on the question why 
> the circle does not lock to inside of the ring.
> I have attached my source code and all images.
> Thank you very much!
>
> Best regards
> Amar
> ------------------------------------------------------------------------
>
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at: http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>