developing clusters in a 3D image

I am very new to MATLAB (and programming) and I'm probably out of my skill level in attempting this, but...

I need to write a script that can take a 3D image and given an x,y,z coordinate within that image test all adjacent coordinates to the given coordinate to see if any of them are greater than 0. If an adjacent coordinate is greater than zero I need to repeat the aforementioned process for this coordinate as well until all coordinates > 0 and branching from the original coordinate are exhausted. Could anyone perhaps give me any hints as to a basic method (not necessarily efficient) to accomplish this? I suspect this process has to be recursive?

Thanks

[Image Analyst]

What do you want as an output after you're done doing the region growing (which is what it's called, not clustering). And it can be recursive but can be done in a single line of code by calling bwconncomp() or bwlabel().

Answer 1

[Matt J]

You could do this pretty easily with bwconncomp()

   BW=Image>0;
   BW(x,y,z)=true;

   CC=bwconncomp(BW);

   lidx=sub2ind(size(BW),x,y,z);

Now just loop through all CC(i).PixelIdxList and see which region contains lidx.

[C.]

Basically I am trying to find ROIs in an fMRI image using previous data about where these ROIs are (The coordinate is the max intensity voxel within an activation cluster). The script will start at this voxel and find all active neighbors until they are exhausted and produce a binary image with 1s in active voxels and 0s elsewhere. I'm a little confused by what the last line is doing, however.

lidx=sub2ind(size(BW),x,y,z);

[Matt J]

It's converting the x,y,z coordinates to linear indeces, which is the format that the pixel coordinates are held in in CC(i).PixelIdxList.

See,

http://www.mathworks.com/help/matlab/ref/sub2ind.html

Answer 2

[Image Analyst]

In addition to the way Matt said, you can use bwlabel and ismember

labeledImage = bwlabel(yourArray);
regionIWant = labeledImage(row, col, slice);
outputImage = ismember(labeledImage, regionIWant);

Or you could use imreconstruct to do a morphological reconstruction:

seedImage = false(size(yourArray));
seedImage(row, col, slice) = true;
outputImage = imreconstruct(seedImage, yourArray);