Dicom image segmentation to detect objects with same pixel intensity neighbor

Dear all,

Attached are the series of CT scan images at different body height level that increases from a (belly) towards b (chest). From such images, I am trying to detect outer chest muscle (green curve) in order to generate its 3D computer model. So far, my Matlab code is able to detect the boundary of the green surface if the outlier structure (red) is far from the green surface fig. a). However, if the outlier red structures are close to the green surface, bwboundaries considers the red outlier surface as well. Here, both the organ with red and green surface have same contrast, this makes thresholding difficult. I adjusted imopen/imerode to remove some pixels so that they do not come closer, but deletes the pixel points from the green surface and gives incomplete representation of the region of interest.

What could be the best way to detect the green curves only with no influence from the red bounded structures?

Thank you very much in advance for any help.

Answer 1

[Image Analyst]

If the other tissue is "touching" the green surface, then it will be tough to distinguish them based on intensity and morphology alone. You might look into Segmentation Net but you'll have to have a lot of images where you've hand traced the green curve to train it.

[ArBio]

My apologies Image Analyst for late reply, I was not notified to the reply on this thread. Thank you and yes, I made up my mind to the Segmentation Network for this problem.