[Insight-users] Geodesic active contour shape prior levelset

petin petin75 at gmail.com
Mon Sep 27 04:11:59 EDT 2010


Hi Kishore,

Thanks for your answer.
I improved a lot the segmentation reducing the shape therm in the levelset
equation
so that the weight of the image term is bigger than before, and now the
results are more
precise then before so we are quite satisfied.

Thank for your idea trying several orientation and then find the most common
one, I will try to
see if the speed is acceptable for our application. I thought to solve this
problem (the initial rotation)
by trying to correlate the average shape coming from the training set with
the image.

Anyway do you know of other works or papers that adds the shape prior to the
levelsets or
can you point me to other shape based segmentation technique ?

-Petin
On Fri, Sep 24, 2010 at 6:55 PM, Kishore Mosaliganti
<kishoreraom at gmail.com>wrote:

> Hi Petin,
>
> I have tried using the ShapePriorLevelSet filter before.
>
> It is true that if you eliminate the registration aspect, then the
> final result will be better. This is because of fewer parameters to
> estimate for finding the global optimal set. Giving a registration
> initialization is a good idea where it is possible.  You can also try
> to automate that by specifying different starting orientations and
> choosing the most common orientation observed (via voting etc). That
> may help reduce the registration burden on the user.
>
> The convergence of the level-set is due to various terms including the
> gradient. What you visually see as the object boundary from intensity
> may not be the location of the minima in the second gradient etc. The
> levelset terms are attracted to a gradient basin. Therefore, you may
> be perceiving the difference. The boundary is always going to be
> difficult because of the partial voluming effects of neighboring
> voxels. I think your result is fairly good from my perspective.
>
> Kishore
>
> On Thu, Sep 16, 2010 at 2:55 AM, petin <petin75 at gmail.com> wrote:
> > Dear Itk users,
> >
> > I'm using  GeodesicActiveContourShapePriorLevelSetFilter image filter to
> > segment 2d sagittal
> > images of vertebrae from L1 to S1. We created an average shape as
> described
> > within the
> > itk programming guide using about 50 images for each vertebrae, so the
> > segmented vertebrae
> > were registered ,we did the pca and so on....
> > After the training set has been created we tried the algorithm on some
> > image, the user was
> > able to click on the vertebrae and from this seed point the segmentation
> > started.  We found that the
> > final resul is a lot influenced by the rotation of the object to be
> > segmented, than we modified a bit
> > our user interface so that the user could specify an initial rotation of
> the
> > vertebrae to be segmented.
> > The results we got were quite good, but still there are some cases like
> > those attached where the
> > final segmentation is not perfectly adherent to the vertebrae border and
> the
> > segmented object
> > is not perfectly rotated.
> >
> > Could you please give me some advice ?
> > Any hint would help.
> > Kind regards
> > -Loris
> > _____________________________________
> > Powered by www.kitware.com
> >
> > Visit other Kitware open-source projects at
> > http://www.kitware.com/opensource/opensource.html
> >
> > Kitware offers ITK Training Courses, for more information visit:
> > http://www.kitware.com/products/protraining.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
> >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20100927/756cfaa2/attachment.htm>


More information about the Insight-users mailing list