Well, that's what I got my Ph.D. on. Probably lots of other people did too. So it's not some simple thing where we can just give you 500 lines of code and it will robustly find stenoses. I don't even have that code anymore anyways. You have a lot of work ahead of you. Like I'm assuming you've also chosen this as your Ph.D. project. So I suggest you consult with your professor and collaborating radiologist on a plan. In the meantime, see how others did it by searching PubMed or VisionBib for similar papers:
By the way, since you have the binary image, you can get the widths by getting the skeleton and multiplying it by the Euclidean Distance Transform. See attached demo.
