By using multiple seed points, I was able to get a larger portion of the lumen during the initial segmentation:

These are the initial segmentations of the left and right stents. Three seed points were used for each stent, one for each of the disconnected pieces shown below. I couldn't determine the connection to a blood vessel on one side of the right stent. Since the left stent gave a better initial segmentation, I've been working mostly with that side.

Once again, the issue was that the 'brightest' part of the TOF data lies outside the 'dark' lumen area in the REST data. You can see below a sample from the imgRegTool with the REST and TOF data loaded (the input images are shown, no registration has been performed). The REST data is shown in the red channel, the TOF data in the green channel, and the voxels that were segmented in this slice are shown by the white outline.

Below you can see what happens when I try to use this data as the 'seed' in SNAP. When I preprocess the data, the segmented area all falls outside the white region that allows growth.

After looking more closely at the data sets, and running registration myself, I don't think that the registration is the problem. The images seem well registered, but the brightest areas from the TOF data lie outside the 'empty' areas in the REST data. I don't know if there is an explanation for this having to do with the MRI scanning technology or technique. I did a little reading, and if I understand correctly, TOF imaging is quite sensitive to blood flow velocity, since the brightest images will be generated when blood from outside the slice plane completely replaces the original blood within the repetition time of the MRI. Could these blood vessels have very slow flow? Is there a reason why flow would be highest near the edge of the vessel? Also, is there a reason that a contrast enhancer couldn't be used during the data acquisition?

As far as improving the initial segmentation goes, I'm planning on trying to increase the amount of lumen segmented by increasing the threshold range. Unfortunately, increasing it above the level I've used in the example above adds a lot of 'noise' from other parts of the image (even with the connectedness requirement). I'm thinking of using a fairly low threshold for the initial segmentation, then running some sort of erosion filter to break off the parts that are weakly connected (hopefully breaking off all the connections that allow the segmentation to 'spill out' into other tissue), and then running another connectedness filter using the original seed points to get rid of these newly disconnected pieces. If that works, I may be able to get more of the initial segmentation inside the 'growth' regions in SNAP, and setting the evolution parameters correctly will hopefully get it to contract and expand in the appropriate areas.

RESPONSE:

Lucine and I looked at the data - we guessed your typo, and were able to identify the slice in question on the TOF images. She doesn't have the 3DTSE REST images. But the primary artifact will be on the TOF images. In this case, it looks like the artifact is a small number of slices below the insertion of the graft into the jugular vein.

There are generally two artifacts in TOF images that cause geometric distortion. Because there is a small time delay between encoding the position of the blood in the three orthogonal directions (as much as 2-4 msec), if blood is flowing obliquely, the x,y, and z positions can be recorded at different times, resulting in an apparent shift in the position of a small element of blood. (Typically z and y are encoded at the same time, and x is later by a few milliseconds). For that to cause a shift, the flow needs to be fast and oblique. In this case, the jugular seems to be going mostly in the "z" direction, and so I don't think that is the cause.

The second cause of a possible distortion is fast streaming of the blood and partial volume effects. In this case, looking at the source images, it is clear that the lumen of the jugular vein is patent, but that the graft seems to be disturbing the flow below the graft. It appears that the flow is fastest near the vessel wall at the point of the very high signal. Because the voxels are of finite size, it could appear that the vessel wall is bright right next to the lumen. My guess is this is the cause of this problem.

I could try to look at the images from the web site - I downloaded them but I don't know what software was used to create them. Are they just a flat file or have they been compressed? I am not familiar with the .raw extension.

Interesting images,

Dennis