Since I was planning on trying out several different filter combinations to see if I could segment more of the vessel from the TOF image, I thought i'd give SCIRun a try. Below is a segmentation using the same network I implemented in my program:

And the best result I could get using the ConfidenceConnected filter:

Increasing the threshold eventually leads to a bleedover into surrounding tissue:

Similar for ConfidenceConnected:

I made a few attempts to create a network that would remove the extraneous tissue from a segmentation like the two above, but without much luck. I also tried a simple 'dilate' filter on a partial segmentation, but (unsurprisingly) it lead to runaway snake evolution in SNAP.

Most of the bleedover occurs at the far Z extents of the image, so I was planning on extracting and processing a tighter region around the stent and seeing if I could increase the threshold without the bleedover. However, the ExtractImageFilter wasn't behaving for me in SCIRun, and there were some other issues as well -- I didn't see an interactive way to specify seed points for the ConnectedThreshold or ConfidenceConnected filters, and there wasn't a way to specify multiple seed points through the existing GUI. I'll have to look into it more thoroughly at some other point, but for now I'm back to using my own application. I have an interactive 'region of interest' feature now, but resizing the result so that SNAP accepts it has been giving me problems this evening. I'll finish that up and hopefully have some more interesting results by the end of the weekend -- if the current methods don't pan out by then, I'll start looking at something different. From my experience with SNAP so far, I'm going to have to get pretty close to a completely connected 'seed' segmentation -- it doesn't grow along the vessels very well. I'll adjust the parameters some more and give that a more thorough test too, though.