[Insight-users] Geodesic Active Contours passing over edges

[Ivan Macia]

Dear ITK users :

I have a problem with segmenting the liver in a series of slices. I first
used the pipeline in the ITKSoftwareGuide. When I set a small value for the
distance to the seeds in the FastMarchingFilter (used here to generate a
distance map) I get a quite nice result although the initial zero level-set
is very far away from the edges of the liver. If I use a larger value, that
approximates better the shape of the liver (being careful of not having any
lacks in this initial level-set) it results in controlled lacks in the
segmented region, which is somewhat senseless to me.

I have realised that what is happening really in my case is that the
Geodesic Active Contours are lacking over the first edges they find in the
Sigmoid filter (used here as Feature Image) and are trying to fit to the
second edges they find. This means that if in the Sigmoid the object I want
to segment is white, then, there is a separation in black surrounding this
area and then there is a transition to white again, the GAC try to adjust to
this last transition. The same happens when I try to use other prior
segmentations that adjust better to the liver area instead of using a Fast
Marching filter as initial level-set.

I tried to reduce the values of the propagation and advection terms in the
GAC and increase the Curvature term, but it does not seem to take effect (I
even tried with Advection = Propagation = 0.01 and Curvature = 30.0). How
can I fix this? Is this a normal behaviour?

Sorry if I haven't explained myself clearly and thanks in advance.

Iván

________________________________

Iván Macía Oliver
Ing. Industrial
Ing. en Automática y Electrónica
Asociación VICOMTech
http://www.vicomtech.es
Paseo Mikeletegi 57, bajo
20.009 Donostia-San Sebastián
(Gipuzkoa) SPAIN
Tlfno: (00 34) 943 30 92 30
Fax: (00 34) 943 30 93 93
e-mail: imacia at vicomtech.es

***Member of INI-GraphicsNet***
http://www.inigraphics.net
_______________________________