Hi Evan,
the main issue with the data (not your fault) seems to be that images have a
strongly anisotropic resolution (thick slice).
The combination between small vessels (of size comparable to the resolution, at
least in one direction) and anisotropy
leads to inaccurate image intensity gradients and consequently to weird
"motion" of the level set surface.
You can try resampling your image to a more isotropic resolution using a
high-order interpolation scheme (such as
cubic or b-spline). Usually this helps only moderately, but it's worth a try.
What I've done in these situations in the past has been to use a more
constrained segmentation, such as polygonal surface
deformation after Marching Cubes extraction or even deformable tubes
(centerline + radius).
The main issue lies in the lack of information in one direction, which you have
to make up for someway or another.
Luca
On Apr 5, 2013, at 1:19 AM, Evan Kao wrote:
> Hello Luca,
>
> Thanks for the references. As for the specific problem itself, I have
> attached a couple images (it is an arterviovenous fistula). I am using dicom
> data from cine-MR measurements. When using the suggested settings of [300 0
> 0 1] during level set segmentation, part of the vessel on the right (vein)
> does not appear. Fewer iterations (30-50) seem to do a better job of keeping
> that segment filled, although it still appears under-filled. I have tried to
> get around this by segmenting the two vessels separately, using different
> parameters for each one. The process becomes a balancing act between the
> PropagationScaling and CurvatureScaling parameters (I used [30 1 1 1] for the
> vein, [200 1 1 1] for the artery to get passable results), and the resulting
> lumenal geometry feels a little arbitrary. But more problematic is that I
> end up having to use the Fast Marching method on the underfilled area to
> adequately fill it, which ends up hurting the reproducibility of the geometry
> and kind of defeats the purpose of using level set segmentation over
> something simpler, like simply eyeballing thresholds with vmtkmarchingcubes.
>
> Please let me know if you have any suggestions.
>
> Thanks,
> Evan Kao
>
>
> On Fri, Mar 29, 2013 at 5:43 AM, Luca Antiga <luca.ant...@orobix.com> wrote:
> Hi Evan,
>
> On Mar 22, 2013, at 9:48 PM, Evan Kao wrote:
>
>> Hello,
>>
>> I am trying to segment a particular set of 3D-cine images with
>> vmtklevelsetsegmentation, but the perhaps due to the shape of the vessel
>> and/or quality of the images, the suggested settings of 300 0 0 1 isn't
>> working well. I've tried playing around with the level set deformation
>> parameters, but the resultant shapes feel just as arbitrary as if I had
>> simply used thresholds directly. Can you recommend any guides or references
>> that can explain in more depth (i.e. quantitatively) how the parameters
>> affect the deformation, and/or also provide an explanation of the additional
>> options available in vmtklevelsetsegmentation that weren't covered in the
>> tutorials?
>
> The level set parameters are those of
> itkGeodesicActiveContourLevelSetImageFilter and
> itkGeodesicActiveContourLevelSetFunction.
> http://www.itk.org/Doxygen/html/classitk_1_1GeodesicActiveContourLevelSetImageFilter.html
> http://www.itk.org/Doxygen/html/classitk_1_1GeodesicActiveContourLevelSetFunction.html
>
> The main reference for level set methods is
> J. A. Sethian, "Level Set Methods and Fast Marching Methods", Cambridge
> University Press, 1996.
> but you can find a lot of material around.
>
> You may want to take a look at my PhDThesis (good old times) at
> lantiga.github.com/about/
>
> If you have noise issues, I suggest you to use a non-zero
> featurederivativesigma , which expresses the width of the
> gaussian kernel used for performing partial derivatives for the feature
> image. It's expressed in physical spatial units
> (e.g. mm) and it's roughly related to the minimum scale that you'll end up
> detecting in the image - i.e. the noise below
> that spatial scale is going to be filtered out, thus not affecting the
> evolution of level sets (again, roughly speaking).
>
> Feel free to post a couple of screenshots of the problem you're having, so
> that replies can be more to the point.
>
> Best,
>
> Luca
>
>
>> Thank you,
>> Evan Kao
>> ------------------------------------------------------------------------------
>> Everyone hates slow websites. So do we.
>> Make your web apps faster with AppDynamics
>> Download AppDynamics Lite for free today:
>> http://p.sf.net/sfu/appdyn_d2d_mar_______________________________________________
>> vmtk-users mailing list
>> vmtk-users@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/vmtk-users
>
>
> <levelset_300_0_0_1.png><levelset_underfilled2.JPG><levelset_30_0_0_1.JPG><levelset_initialization.JPG><levelset_undefilled.JPG>------------------------------------------------------------------------------
> Minimize network downtime and maximize team effectiveness.
> Reduce network management and security costs.Learn how to hire
> the most talented Cisco Certified professionals. Visit the
> Employer Resources Portal
> http://www.cisco.com/web/learning/employer_resources/index.html_______________________________________________
> vmtk-users mailing list
> vmtk-users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/vmtk-users
------------------------------------------------------------------------------
Minimize network downtime and maximize team effectiveness.
Reduce network management and security costs.Learn how to hire
the most talented Cisco Certified professionals. Visit the
Employer Resources Portal
http://www.cisco.com/web/learning/employer_resources/index.html
_______________________________________________
vmtk-users mailing list
vmtk-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/vmtk-users
