bwlabel doesn't work as expected and what would be ideal method to segment disconnected regions like this example

Hi,

This is a segmented image of a brain from thresholding segmentation, Now I want to get A, B, and C regions, which are all brain regions.

when I apply some kind of operation. For now, i am using bwlabel to find the blob areas and correspondingly extract bigger or circular blobs of some kind and get blobs A, B,and C. but the problem is when I do bwlabel I don't get A, B blobs.

Here is the code and attached images before and after bwlabel:

temp = before_dicomFiles{1,15};
    %figure, imshow(temp, [LOW1 HIGH1]);
    temp = double(temp);

      newImg = temp; % this is the actual brain image used for thresholding
      newImg(newImg <0 )  = 0; % set all negative pixels to zeros
      newImg(newImg >1080) = 0; % remove anything over this range
      figure, imshow(newImg, [LOW1 HIGH1]); % thresholded image
      BW = bwlabel(newImg,8) ;
      figure, imshow(BW, [])  % Labelled image missing A, Blob areas

Suggest me some methods to trace back regions A, B and also better segmentation methods to not miss all the brain regions

Thanks, Gopi

Answer 1

[Image Analyst]

Try this:

binaryImage = temp > 1080;
% Fill holes (depends on skull having a hole through it.
binaryImage = imfill(binaryImage, 'holes');
% Extract 3 largest blobs.
binaryImage = bwareafilt(binaryImage, 3);
labeledImage = bwlabel(binaryImage);
props = regionprops(labeledImage, temp);

[Gopichandh Danala]

Thanks, it worked, but lets say all the required blobs are not the top 3 blobs so, in that case can i give a manual boundary and ask the program to pick the blobs which are in the boundary regions drew by the user

[Gopichandh Danala]

The reason for this the above series of image are of a brain and it is easy to segment the region from the centre images where the whole brain is a single connected blob, but as it goes towards edge slides in series it disconnects and forms 2, 3 blobs as in above picture, so i am thinking of segmenting center slide where blob is of max area and then use its boundary and may be map all blobs within its region for the sequence slides , Any suggestions about this idea?

[Image Analyst]

You might try using bwconvhull to get the convex hull. Then use imerode() to shrink it down. Then use that as a mask to "erase" the outer parts (skull, etc.);

mask = bwconvhull(binaryImage);
mask = imerode(mask, true(15));
binaryImage(~mask) = false;
% Then the previous code I gave.

[Omkar Patil]

Thanks, I will try this and give you an update next time I go to lab, Oops i am commenting from my friends computer

Answer 2

[Saurabh Gupta]

You can use the Image Segmenter App to perform the segmentation, then export it as a MATLAB script to observe the methodology used to implement further tasks in an automated manner.

[Gopichandh Danala]

In real time this kind of apps are pretty useless and thanks for your suggestion