[Insight-users] Level sets Chan-Vese

[Uday Kurkure]

Hi Brad,

I tried to experiment with the different implementations.  I
experimented with a synthetic image and a real image (which
unfortunately I cannot share).

In attachments you can see:

- a.png : synthetic image (100x100)
- abw.png : initial mask

I experimented with 3D synthetic also which was constructed by
concatenating the 2D image (100x100x100). Here are the results:

- outDense2Da.png :
output of dense levelsetsv4.
Perfect segmentation on 2D and 3D synthetic image.
[On my real image the mask didn't grow at all to connect two regions
but it did remove the background from the mask.]

- outWhit2Da.png :
output of Whitaker sparse levelsetsv4.
Erroneous segmentation in 2D and 3D. There are many extra pixels
included in the segmented mask especially near the corners.
[It works as expected for my real image in 2D. But in 3D real image it
gets stuck for hours at converting the initial mask to sparse format.]

- outDenseReview2Da.png : output of dense levelsets in review folder.
The segmented mask didn't include the boundary pixels (something to do
with thresholding of level set function?).
[On real image the mask didn't grow at all to connect the two regions
but it did remove the background from the mask.]

- outWhitReview2Da.png : output of Whitaker sparse levelsets in review folder.
Perfect segmentation in 2D and 3D.
[On real image the initial mask doesn't change much. Didn't even
remove the background.]

I am still trying to figure out the issue with the real image vs
synthetic image.

I hope someone can explain the different behavior of these
implementations on the synthetic image.

Overall, for synthetic image 2D/3D, the dense levelsetsv4 and sparse
Whitaker in review perform accurately.
On 2D real image, sparse Whitaker levelsetsv4 performed best but in 3D
it gets stuck at converting the initial mask to sparse level set
format.

Any help/suggestions/comments are appreciated.

Regards,

- Uday







On Fri, Feb 15, 2013 at 8:25 AM, Bradley Lowekamp
<blowekamp at mail.nih.gov> wrote:
> Hello Uday,
>
> I Arnaud Gelas has contributed a lot of the work on Chan-Vese level-set. I was hoping he would have chimed in by now.
>
> I have not heavily used either of these level-set so I'll do my best. The ITKv4 level-set framework is very flexible with the abilities to create new terms, try different level-set representations, and provide a unified framework for dealing with multiple phases. This has some performance compromises for large single level-sets but it's my understanding it works quite well for a large number of smaller level-sets in a large volume. There is also some nice visualization tools that are in the ITKv4 repo too.
>
> I have used the review filters my self with some success. And it's my personal preference for the type of data I have used. However, when I tries to wrap it for SimpleITK I generated some seg-faults with the parameters I was providing.
>
> I'd recommend trying both and letting us know how they work out for you.
>
> Brad
>
> On Feb 12, 2013, at 3:28 PM, Uday Kurkure <ukurkure at uh.edu> wrote:
>
>> Hello all,
>>
>> It seems there are two implementations of Chan-Vese in ITK in folders
>> - LevelSetsv4 and ITKReview.
>> Which one is recommended to use in C++ code?
>> Are these going to be merged together at some point in future?
>>
>> In python I guess only the filters in the review folder are available
>> after wrapping.
>>
>> Regards,
>> - Uday
>> _____________________________________
>> 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.php
>>
>> 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 --------------
A non-text attachment was scrubbed...
Name: outWhitReview2Da.png
Type: image/png
Size: 265 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130215/b999d54d/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: outWhit2Da.png
Type: image/png
Size: 303 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130215/b999d54d/attachment-0001.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: outDenseReview2Da.png
Type: image/png
Size: 295 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130215/b999d54d/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: outDense2Da.png
Type: image/png
Size: 265 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130215/b999d54d/attachment-0003.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: abw.png
Type: image/png
Size: 145 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130215/b999d54d/attachment-0004.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: a.png
Type: image/png
Size: 165 bytes
Desc: not available
URL: <http://www.itk.org/pipermail/insight-users/attachments/20130215/b999d54d/attachment-0005.png>