Segmenting objects of similar intensity from an image

Wednesday, July 9, 2014

I am working on segmentation of medical images (specifically CT scans) in Python and C++. I have managed to successfully eliminate bone, air, and fat from the slices using a combination of thresholding, region growing, and morphological operators. The main problem that is left is segmenting vital organs from muscles, since they have very similar intensities and often abut one another.

For example, in the image below, a piece of bowel is seen abutting abdominal wall muscle:

enter image description here

The desired segmentation result is as follows:

enter image description here

Another acceptable segmentation is as follows:

enter image description here

My question is: what segmentation algorithm should I be using to achieve one of the desired segmentation results? So far, I have tried:

  • Thresholding: does not work because the two regions are too similar in intensity.

  • Region growing: does not work because the regions to be segmented are "connected".

  • Watershed transform: results in over-segmentation and does not accurately capture the boundaries.

  • Active contours: fails to find an appropriate boundary between the two objects.

Any pointers are appreciated.