Distinct peaks for fat and gland tissue pixels

I have the histogram of a 16 bit dicom mammographic image that contains pixels of gland and fat tissues . I am using log sliding threshold method to seperate them. But the pixels of fat and gland tissues are not seperated completely , peaks are not clear . The Peaks are almost flat and there is no clear valley between them (it is almost flat). So I can not select the threshold position. how can I identify these two pixel sets and get a clear valley between the peaks?

Answer 1

[Image Analyst]

Sees like a medical question. Use a different kVp on your x-ray system perhaps?

So the fat and glands cannot be solely determined by gray level so that for a certain gray level, it might be either. Maybe use shape information?

Have you checked the literature on this: http://www.visionbib.com/bibliography/contentsmedical.html#Medical%20Applications,%20CAT,%20MRI,%20Ultrasound,%20Heart%20Models,%20Brain%20Models

[Ruhul Amin]

Dear Image Analyst

I have checked the literatures but none of them is similar to my problem it seems. Yes you are right , a certain gray level could be either for gland or for fat tissue pixel. That may be the reason why it comes flat or like straight line in some parts. To get valley or difference in between them , I guess, if I can reduce the intervals in x axis scale or in gray level scale , it might show the valleys . Y axis is already log scale and I am using your template of sliding threshold bars. So please help me- how can I change the scale intervals in x axis or in other words compress the gray levels ( 65535 levels , as it is dicom image) in x axis of histogram?

[Image Analyst]

You can divide your image by 256 to bring it down from 65536 gray levels into 256 gray levels but that won't solve the problem.

You may have to write your own custom threshold finding algorithm that is customized for your particular histogram shapes. However, it may be that there is no single threshold that gives a good segmentation - maybe any/every threshold is unacceptable. In that case you will have to take into account some other features like shape or things like that.

[Ruhul Amin]

thanks a lot. I will try for that