Blob for tumor Segmentation

Hello, I'm working on segmenting and seperating the tumors from Brain MRI scans. I uses Gaussian filter to pre-process,and then i used thresholding( I >100) to threshold my image. I then used morphological operations to smooth the image and bwlabel to label and then to get the boundaries of the labels. I want to seperate the inner blob\circular region so that I can seperate the tumor from the images. Pleaseee help me with this

What im finding common in almost all the images is the fact that the tumor is of circular shape, and i need a method to somehow extract that and discard all other aspect as for now. Any other methadology is preferred as well. Thank you

Attaching the picture of results as ref

Answer 1

[Image Analyst]

See my skull stripping demo. It does this and gets rid of the skull by finding the bright skull and then eroding it away.

Your decision/assumption that all tumors are round is not a good or robust one. For better algorithms, see PubMed or VisionBib:

http://www.visionbib.com/bibliography/contentsmedical.html#Medical%20Applications,%20CAT,%20MRI,%20Ultrasound,%20Heart%20Models,%20Brain%20Models

If it's just a simple homework problem, fine, but if you're working on your Ph.D. you'll have to come up with something better than all the prior published algorithms (none of which are as simple as my code or what you proposed - they're much better).

[Muhammad Hadi Qureshi]

Thanks a BUNCH! Well, not a research topic, its my Bachelors level project:)

Btw, could you please assist me a bit more. Im unable to understand how to select the threshold value for skull stripping. And how have you used histogram to determine that value. Even if i opt for 200 threshold value, my output image gets all black. :(

Would be reallyyy glad if you can assist in this. TIA!

[Image Analyst]

See my interactive thresholding app in:

Image Analyst's File Exchange

Using that you can adjust sliders to determine what threshold looks best.

[Muhammad Hadi Qureshi]

Great. Thanks! Its working well now. You assisted a lot and it means the world!

Im able to detect homogenously enhancing tumors. But, Im having trouble with these kind of hetrogenously enhancing tumors...if i apply threshold, theyre removed. Could you please assist in this one

[Image Analyst]

It could be a Ph.D. project, and I'm sure it has been. So I can't devote that much time to you. All I can do is point you to existing articles in PubMed or VisionBib (USC Annotated Computer Vision Bibliography) where people who have studied that have published their methods. No one here will give you a better method than you can find there. It's not trivial. Nothing that can be solved in 5 minutes or less. It will take, literally, years. And I'm sure they're still working on improving the algorithms.

[Muhammad Hadi Qureshi]

Thanks:)

Means a lot!

[Muhammad Hadi Qureshi]

Also, please let me if i shall cite your code, as im using your code for skull striping and extracting the tumor?If so, could you please guide me the procedure how to?

[Image Analyst]

No need to cite but you can put my name and this URL in your source code if you want. I don't have a paper on this since it's to basic/elementary to be worthy of a publication. It's just common sense stuff.