Brain extraction using Kmeans and connectivity.

Hello, I'm trying to perform a brain extraction of a slice from a NIFTI volume. I wrote a k means segmentation code, and that way I have gotten as far as:

and now I am trying to use connectivity, based on this tutorial but I'm not sure if something similar can be done if the parts I want to extract are not in the borders. In the image, the brain is separated from the skull and there's an empty region between them. Is there a way of separating these regions? I tried something like this but got stuck after I got the connected components structure.

Thanks in advance

[Image Analyst]

They are separate. They're not touching or connected by a path of the same gray level. There's a dark ring between them. Please describe exactly what you want, because they look separated to me. What does "separate" mean to you?

[Fer]

Yes, what I want to do, if it's possible, is take advantage of that ring and the fact that they're separate, and remove the outer part, keeping only the brain.

[Kirby Fears]

How many connected components did you get back from bwconncomp ()?

You should be able to get each one by accessing the PixelIdxList field.

cc = bwconncomp(bw, 4)
grain = false(size(bw));

% instead of 50 try 1, 2, 3, etc :
grain(cc.PixelIdxList{50}) = true; 
imshow(grain);

[Fer]

I did, I actually got as far as this:

However the label corresponding to the skull is not always the same. Running the same code on the same image will sometimes result in the segmentation of a different structure.

I would like to come up with a technique that's reproducible for different images. Any suggestions on where to look into this??