[ITK] CT liver segmentation

[Jan Witowski]

Hi Dženan,
Thank you very much for your reply. I really appreciate that.

I can definitely see my mistakes right now:

1) When I was looking at the output of the watershed filter, I passed it through LabelToRGB() - don’t really know why. That’s the reason why I’ve been using wrong thresholds.

2) Also, I am using Fiji (I had some issues running ImageJ with latest Java+OSX El Capitan) to display my images, and the axis there are marked in microns instead of pixels. That also caused some trouble, because it made me place wrong seed locations.

Once again, thanks a lot for your kindness. Code after your modifications works flawlessly. :)

Best regards,
Jan


On 28 grudnia 2015 at 17:07:55, Dženan Zukić (dzenanz at gmail.com<mailto:dzenanz at gmail.com>) wrote:

Hi Jan,

the code was fine, you missed to match the seed position to the value coming out of the watershed filter (imgWS) in that location. As you were feeding the output of the watershed filter to the segmentation filter, you needed to give the thresholds related to the values of that image and not the original input image (imgSeries).

My modification is attached. ImageJ connection did not work out of the box for me so I wrote intermediate images to disk in order to inspect them.

Regards,
Dženan

On Sun, Dec 27, 2015 at 4:49 PM, Jan Witowski <jan.witowski at student.uj.edu.pl<mailto:jan.witowski at student.uj.edu.pl>> wrote:
Hello Dženan,

Thanks for your assistance.
Full code of my example is located here: https://gist.github.com/jwitos/9abe7124e28ce2e31b6e
I’ve also uploaded the DICOMs I’ve been using (30~MB, 152 .dcms inside) if you want to test them out: http://cm-uj.pl/files/cleaned.zip

Best,
Jan

On 27 grudnia 2015 at 17:35:24, Dženan Zukić (dzenanz at gmail.com<mailto:dzenanz at gmail.com>) wrote:

Hi Jan,

that sounds like some silly simple mistake. If you provided a short compilable example with the accompanying input, somebody will probably be able to point out the problem.

Regards,
Dženan

On Sun, Dec 27, 2015 at 10:20 AM, Jan Witowski <jan.witowski at student.uj.edu.pl<mailto:jan.witowski at student.uj.edu.pl>> wrote:
Hello,

I am working with CT abdomen images and trying to segmentate the liver (liver itself, vessels and the HCC tumor). I managed to do it using OsiriX which was rather manual segmentation, and now I am working to do the same using more accurate and semi-automatic algorithms with usage of SimpleITK. I don’t really have much experience with ITK and image manipulation which is why I ask you guys for some help.

Currently, I want to try the watershed method. I loaded my DICOM series, ran the GradientMagnitude filter and proceeded with MorphologicalWatershed (using arg Level=17, value chosen manually). I got pretty nice output image, I guess:
http://jwitowski.com/content/images/2015/12/WSsegmentation-4469-0.jpg

But now, the question is - how do I take that liver out of the image above? My best guess is, I should run a region growing segmentation.
I tried to do that, but - no matter what parameters I choose - I always get the all-black image as an output. The whole image is black. I tried different thresholds and other starting seeds, but… no result.
Here is a gist of my segmentation part:
https://gist.github.com/jwitos/4b953b3ef873fee3965c

I would really appreciate any help or tip


Best,
Jan Witowski
Collegium Medicum Jagiellonian University in Krakow

_______________________________________________
Community mailing list
Community at itk.org<mailto:Community at itk.org>
http://public.kitware.com/mailman/listinfo/community



________________________________
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20151228/bda279f7/attachment-0001.html>