[Insight-users] Re: Left Ventricle Segmentation, MR data

[Luis Ibanez]

Hi Christian,

It is unlikely that you will be able to get
closed contours that are simultaneouly thin.

You should probably approach this problem in two
stages.  First get a topologically consistent
segmentation. Second use this first segmentation
as initialization for a second method that will
take care of producing a geometrically correct
segmentation.

Since GAC works for solving the first stage,
you only need to look for a methods suitable
for the second stage.

You could look at things like the conditional
dilation methods in mathematical morphology.


---

The computation of level sets will not take
spacing into account. This means that the
front propagate as if the voxels were cubic.
This will be unfortunate in very anisotropic
images because the front will move faster
in the out-of-plane direction than in the
in-plane directions.


Using stopping planes is certainly a good
approach for driving the evolution of level
sets.  Please look at the SNAP application in
InsightApplications. This program implements
stoppers in order to facilitate the control
of leak during the evolution of the level set.




    Regards,


       Luis



---------------------
Christian Dold wrote:
...
> Now I am segmenting the left ventricle of the heart based on MRI data
> and have some trouble to generate a good Input level set for the GAC
> algorithm. Please have a look to the edge Image with an avi file reader.
> The problem is we often get open contours. To solve these proble we
> producing big edges, otherwise the GAC will not work. The problem is no
> we lose the accuracy in calculating the volume of the left ventricle,
> means we need closed contours with small edges!
> 
> I use only one seed point. The segmentation pipeline used is shown in
> figure 9.18 (ITK Software Guide). As Input for the Fast Marching the
> edge image is already used.
> How can we get smaller egdes and a closed contour?  We played a long
> time with the values of the sigmoid no chance!
> 
> Further we will do an 2D oblique cut in the plane of the mitral valve to
> segment only the left ventricle and not pieces of the left Atrium too.
> At the moment we are using the XYZ ROI to minimize the dataset to a
> smaller cubic size. Does the algorithms works if we dont have a cubic size?
> I dont think so but is there is a possibilty to make like a stop 2D
> cutting plane  for the algorithm.
> 
> Thanks a lot
> Christian
> 
> 
> 
> ------------------------------------------------------------------------
>