[Insight-users] bad segmentation result of fastmarching method!why?

[Luis Ibanez]

Hi Xujf,

It is hard to tell without looking at the images...

but,
it is natural that an image with thinner edges and
discontinuous edges will be more prone to leaking.

You could fine-tune the parameters of your speed image
in order to try to decrease the speed in the places
where you are getting leaks.

If that doesn't work, you could simply move ahead
and use the ShapeDetection level set that takes into
account the Curvature of the surface, or the
GeodesicActive Contour Level set that takes into
account the Cuvature and also uses an Advection term.

These two LevelSet methods are capable of preventing
leaks. You should fine-tune the relative scaling of
the Propagation, Curvature and Advection terms.

Detailed descriptions of these LevelSet methods are
available in the SoftwareGuide.

   http://www.itk.org/ItkSoftwareGuide.pdf

Section 9.3, pdf-page 363, paper-page 337.

---

You may want to play with the ITK plugins available
in VolView. The free version of VolView can be downloaded
from:

    http://www.kitware.com/products/volview.html

You will find plugins for FastMarching and GeodesicActiveContours
among others.



Regards,



   Luis


----------------
xujf wrote:

> Hi,all:
>      I used fastmarching method (itkfastmarchingimagefilter) to segmentate a 3D iamge(call it iamge A),and the 
> segmentation result was very good . Today I use the same programe to segmentate another 3D image(call it image B)
> ,but get a bad result in which the front runs out of the boundary.
>      image A and image B are similar and are both brain image. image B is a little smaller and brain-box is 
> much thinner(in fact in somewhere it is not continuous).I think what causes my wrong result is just the thin
> brain-box. Is my judgement right?how can I solve this problem?
> 
> Thanks in advance!
> xujf