How to detect the edema robustly

Dear Sir/Madam

I have to process the attached image, and I should detect the white part in the black contour(edema) automatically, and it is just one of images which should be detected! My code is

im2bw(image,.4);

but I don't think it is robust, because it depends on the graylevel of the edema. Could you please give me some methods to solve this problem? Thank you very much :-)

This is what I got after processing:

[Star Strider]

Extravascular fluid collections can either be transudates (specific gravity < 1.015) or exudates (SG >1.015). Because they have different compositions, they would produce different CT or MRI densities. You need to allow for this in your code.

[Neil Liu]

Thank you for giving directions to me~!

[Star Strider]

My pleasure!

Nothing in medicine is ever straightforward. That also goes for you later question about pericardial effusion. The radiographic density entirely depends on the aetiology of the effusion. The best you can hope for is to somehow standardise the myocardial density and standardise tbe pericardial density, and look for whatever is between them.

Answer 1

[Image Analyst]

Take the histograms and manually threshold. Maybe you'll see some pattern where the best threshold lies.

Maybe you'd like to try the Frangi filter on it http://www.mathworks.com/matlabcentral/fileexchange/24409-hessian-based-frangi-vesselness-filter

Or a cosfire filter: http://www.mathworks.com/matlabcentral/fileexchange/49172-trainable-cosfire-filters-for-vessel-delineation-with-application-to-retinal-images

[Neil Liu]

Thank you for answering me! Actually I want to solve this problem automatically, I mean, not try to use the threshold again and again, because my aim is to put these codes on different images.

[Image Analyst]

Every image processing problem where you end up measuring something (e.g. size, shape, area, intensity, etc.) at some point gets down to thresholding. Sorry, but that's just the way it is. If the thresholding is done by some algorithm instead of you doing it manually with my awesome manual thresholding app then what's wrong with that?

[A-J KHAN]

@neil liu hi..... Can you plz share the start of code, because i can't understand what are u saying..

I suggest morphological filter and then Varitional segmentation method.