What is the best method for color and shape-based segmentation for this particular case?

1 view (last 30 days)
The project I'm working on involves recognizing cancerous tissue on a CT scan based on its shape and color (circular and white) To explain the idea to my professor, using Paint I made some kind of "mock" CT scans of a healthy (H) and a cancerous (C) person. By comparing the images with difference I isolated any structure C had that H was lacking, and then using imfindcircles I got all the circle shaped structures.
This time I'm trying to analyze real CT scans individually and find the circle-shaped cancerous tissue by pattern recognition, but here is where I realize my knowledge in Matlab is very little because I'm not sure what method should I use.
1. I first thought about using K-means clustering, but my code just won't separate the image for the life of mine. Could it be because the image, since it's a CT scan, has basically just variations of two colors? (black and white)
2. I've been trying with Color-Based Segmentation Using the Lab* Color Space, folowing this tutorial . Doesn't work so far as well, but I'd like to now if it's a valid approach. Once I can have in an image only the white and bright portions I want to test with imfindcircles and hopefully isolate the cancerous tissue.
3. Another option is to follow this other tutorial. Would it work? A real CT scan of a cancerous lung looks like this , I'm not sure if the contrast between cancerous tissue and surroundings is strong enough to use this method.
Can you give me some feedback or some advice? I can't stress enough my lack of experience with Matlab, this is the first time ever I'm using it for a project but I''m trying my best. Thanks!!

Answers (1)

Image Analyst
Image Analyst on 27 May 2015
You forgot to attach your images.
CT images don't have color, unless the gray scale image has been pseudocolored but that's not what you deal with. If someone pseudocolored it then that means they already segmented it. So you're dealing with a gray scale image, and just forget everything anyone told you about using color.
There is no way you can subtract a CT slice of a healthy person from a cancer victim and have it find the cancer. Not sure if anyone told you that, but just put that out of your mind. The anatomy just won't line up.
They've been working on this problem of tumor detection for 40 years and will still be working on it 40 years from now. Go here to browse the papers on it: http://www.visionbib.com/bibliography/contentsmedical.html#Medical%20Applications,%20CAT,%20MRI,%20Ultrasound,%20Heart%20Models,%20Brain%20Models Pick one and code it up. It's not going to be as simple as finding circular, high contrast regions. If you're just doing some homework problem or undergrad project with a plastic phantom, fine - do something fake but easy, but if you're a professor, or a grad student, or in a medical imaging company, that just won't be sufficient.
  2 Comments
Maira C
Maira C on 27 May 2015
Edited: Maira C on 27 May 2015
I doulecheked the images I intended to put and they have an hyperlink to imgur. Can you explain me what do you mean with this?
"There is no way you can subtract a CT slice of a healthy person from a cancer victim and have it find the cancer. Not sure if anyone told you that, but just put that out of your mind. The anatomy just won't line up."
Do you mean it's impossible to determine if a person has cancer or not by comparing their CT scans with those of a healthy person? If that's what you meant, I also realised that and that's why I'm trying to do an individual analysis.
I found this paper that is very similar to what I intend to do. The authors also partitioned the CT scan by image segmentation, and I guess my question is more along the lines of: which image segmentation method is the best for the situation I explained above?
I've been given this topic to write my undergraduate thesis, and I'm on the early stages of trying to figure out exactly what and how to do it. I'm not well aware of the limitations and challenges I may face with this topic, so thanks for the insight and your answer!
Image Analyst
Image Analyst on 28 May 2015
Not sure what you want explained. CT images are not color images. If you have one specific image that definitely is known to have a circular blob in it, then yes, you can find that. That is a lot easier than doing it in general for any arbitrary CT image.
If you have a tumor, then we cannot find your tumor by subtracting a CT slice of me, from your CT slice. This should be obvious and I think you know this now and edited the original question to get rid of mention of subtraction.
It looks like the paper authors used MATLAB. Ask them for their code.

Sign in to comment.

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!