I'm Dennis Parker, I'm a professor in the department of radiology at the University of Utah, and I appreciate your interest in our poster, which is entitled...... The Application of the Depth Buffer or Z Buffer Segmentation Algorithm in Magnetic Resonance Angiography, with applications also in computed tomography. And this algorithm is one that's based upon the maximum intensity projection (MIP) algorithm, which is used in the many MR and other applications for the display of blood vessels. What we've found is, if we look, not at the maximum intensity projection image, but if we look at the Z buffer of the maximum intensity projection image, what you see is an image where the vessels look very smooth, and the background itself looks very rough.

And this is a property that was new to us, when we noticed it the first time. And we found that if we looked, if we wrote an algorithm that simply looked for smoothness, that we had an algorithm that was a very strong segmentation of the vessels from the background, because the background in the Z buffer image tends to be very rough. And the vessels themselves tend to look very smooth. The algorithm is based upon doing a least squares fit, which is a ... which is a mathematical technique for finding smoothness, or fitting a line through vessels, or fitting a line through points, in ... in this case, in the two dimensional space.

We try and fit this line, and the measure that we use, is the goodness of fit, or the chi square, which is just simply, a measure of how close the line itself fits the points that were actually measured, or actually in the Z buffer.

And by doing this in four principal directions, around every point, we take the minimum chi square, and that value is a very low value any place where there's a vessel. And it's a very high value in the background. And that gives us a very very strong segmentation of the um, the blood vessels from the background.

And, at that point, we apply connectivity; just simply connect points, which are next to each other, in the um, in this chi square image, that are also close in Z. And we get um, an image that shows a structure which ... where the blood vessels tend to be grouped. And they tend to be very well segmented from the background.

The final step of the algorithm is just simply to um, map the segmented structures back to three dimensions, and apply region growing and we end up with a ... a very nice three dimensional representation of anything that appeared in the maximum intensity image. The ... the strength of this algorithm is that it's extremely robust, in segmenting the blood vessels or anything that appears in the maximum intensity projection image, from the background.

We've also applied it to contrast-enhanced images of the heart, and the descending aorta for magnetic resonance. And it tends to work in almost every case. We're hoping to do continued research on this algorithm. We're applying this algorithm to the problem of segmenting arteries from veins. It turns out this algorithm tends to work in any situation where you're doing a projection from multiple dimensions, and selecting a single point, or a single group of points.

Back to main page