Automatic Segmentation of Mass in mammography images

I want to use level set segmentation by selecting the ROI in the Mass tumor area not whole the image. The image attached here, I marked Blue line, the tumor area , this is the area that I want to automatically segment. Any idea please and thank you very much

Answer 1

[Image Analyst]

I'm not sure how level set method can be used here, at least from the description of it given in given in Wikipedia. Do you have a reference? If not, I'd first try simple thresholding since it looks to be brighter than the surrounding tissue. If that doesn't work I'd try activecontour, like in the attached demo, or maybe try some morphological methods, or just look in the literature for successful methods that have been published already. You're not the first one to try to find anomalous masses in mammograms, so why not use methods already tried and true?

[Adnan Ibrahim]

Thank you very much

[Ahmad Fouta]

I applied the code in the active contour demo, and this the image i get: </matlabcentral/answers/uploaded_files/41428/applying%20contour.png>

what modification should I do to highlight the inner mass within the mammogram. and how can it be extracted? Please advice

Answer 2

[Nitin]

You might want to have a look on Active contours/snake. This site has a nice tutorial including source code in Matlab.

[Anand]

By the way, the Image Processing Toolbox already has active contour based segmentation. Look at the activecontour function.

If you could attach the original image, that would be helpful.

[Adnan Ibrahim]

Thank you I got it.

[Warid Islam]

Hi Anand,

I tried to use the activecontour function on my image to detect tumor. The brightest part of the image is the tumor. But I am not getting any result.

Answer 3

[Faraz]

Bro, that's not the tumour area and this is not how you do it. This is fibroglandular tissue (with possibly tumour hiding somewhere, or possibly not) may be BIRADS 3 mammogram. Its the breast region you are trying to segment? But if you think you are segmenting mass here, you are wrong!

You may want to try superpixel based SLIC segmentaion instead of levelsets etc. , google it .. there are a number of implementations available (e.g vlfeat.org).