How to change pixel values of background of an MRI Segmented result ?

Hi, I am trying to run a script for object segmentation based on GMM. As the example presented below the image is segmented into 3 classes. the problem is that the background of the image has the same pixel value of one of the segmented classes. If i change into 4 classes the segmentation result is worse. Is there any solution?

Answer 1

[Walter Roberson]

imclearborder -- unless the white part touches the border (if it does, then pad the image with a row of background before doing the clearing)

Answer 2

[Ynne]

another example with this image, the black region within the brain is a class. So when i want toextract it from image it will be fused with the background of the image (because they have the same pixel value 0 in this case) Is there any solution to extract this class without the background ?

[Walter Roberson]

Once you have cleared the border to black, you can imfill('holes') . That will get you an ROI mask.

The median of the outside edge values is probably going to be the border level. Pad the image with a row of that in case the real stuff touches the edge. imclearborder. imfile('holes')

But as for extracting: what data structure does your GMM segmentation need? Does it already handle being passed an ROI mask? Does it just want a vector of values? If it wants something 2D then you could extract down to a bounding box but you would still have problems.

Answer 3

[Image Analyst]

What function are you using to get the GMM model? gmdistribution() or fitgmdist() followed by cluster? You can make a vector of pixel values to assign a cluster label to. Then you make another N-by-2 array that has the locations (row, column) of all the pixels in the ROI. Then with that you can create a classified image, and assign the pixels outside the ROI to some class number that identifies them as not assigned, like 0 or -1 or whatever.