He was in the right place at the right time. 

It was 1996. Stephen Aylward, a graduate student in computer science, needed an assistantship. Elizabeth Bullitt needed an assistant. 

Bullit, associate professor of neurosurgery, had a problem. She wanted to be able to use computer technology to view individual vessels in the brain in a three dimensional (3D) format. Traditionally, surgeons identified trouble spots, such as aneurysms, by using X-ray angiography-a dye, or radioactive tracer, is injected into the bloodstream, making the blood vessels visible when x-rayed. While the blood vessels show up clearly, they are only two dimensional (2D), which makes it difficult to tell if there are other blood vessels in the way. 

There’s also magnetic resonance imaging (MRI), which produces 3D images of the brain, but the drawback here is that the resolution is very low. “It’s like looking at a big plate of spaghetti,” Aylward says. “The vessels are all overlapping one another and twisting and turning about.” 

So Aylward, who also had an interest in medicine, was put to the test. His task was to produce a 3D image of the brain so individual vessels could be seen more clearly. 

Within two weeks, Aylward, with help from Stephen Pizer, professor of computer science, had a software program up and running that could do just what Bullitt had asked for-extract individual vessels in the brain, so they could be viewed individually or as a group from any point of view. 

Bullitt was pleased. “Aylward just came out of the woodwork,” she says. “He’s phenomenal.” 

Since that first collaboration, the software has really developed. And Aylward has been there to see it all through-as a graduate student, as a faculty member, and now as a co-principal investigator on a recently submitted grant proposal. 

After graduating in 1997, Aylward accepted a joint position as an adjunct assistant professor in computer science and a research assistant professor of radiology. At that moment Aylward became a novelty. It’s not often that someone first works as a graduate student and then as a faculty member on the same project. And, he was a computer scientist who had jumped the boundary to work in radiology. 

But Aylward wouldn’t have it any other way. “The best part for me,” Aylward says, “is that I’ve had the opportunity to improve the software-the speed, accuracy, and ease-of-use of the program.” 

Now that many of the finishing touches have been put on the software-called VTree 3D because the individual vessels look like vascular trees once they have been extracted-Aylward and Bullitt are putting it through the wringer. As coprincipal investigators, they’re trying out the software on radiologists, who will test the software by looking at the different images it produces and seeing which ones are most useful. 

One of the advantages of this software, Aylward says, is that it can be run on just about any PC. “In fact,” he says, “I did most of the development on a laptop computer.” 

Another handy feature is that surgeons can extract all of the vessels in the brain in about 30 minutes. “Often times surgeons are quite nervous about performing a procedure because they cannot visualize these things well. We feel that this software helps to alleviate most of that fear,” Aylward says. 

Using the software, surgeons look at a two-dimensional slice of an MRI and point-and-click on a vessel of interest. The software then automatically finds the middle of that vessel, determines its width, and displays that vessel in three dimensions. This allows doctors to look at vessels of the brain, lungs, or kidneys in three dimensions as opposed to mentally forming a 3D image from multiple 2D slices. A vessel can also be highlighted in a color of the user’s choice to make it easier to tell vessels apart. 

“Surgeons need to be able to visualize everything that’s going on.” Aylward says, “If you’re going to approach an aneurysm, it’s important to know things like where its supplying vessels are and what arteries are around it. They also need to make sure there aren’t any other vessels blocking the way when it comes to surgery.” 

Right now, Aylward and team believe their technique is one of the best, considering its completeness and ease of use. They are even considering licensing the software to a company so that it can be used around the country. 

While Aylward continues to work on the project he helped get off the ground in graduate school, he’s also been able to stake out an area of work as his own. “It’s very hard to be a faculty member at the same university you attended as a graduate student,” Aylward says, “because the faculty still view you as a graduate student, and you still view yourself as a graduate student.” 

While Aylward’s work is still part of MIDAG, the Medical Image Display and Analysis Group, he’s now focusing on computer-aided diagnosis-helping radiologists detect and diagnose diseases. 

One of those diseases is breast cancer. While mammography has helped doctors detect breast cancer with more accuracy, there are still many bugs to smooth out. 

With funding from a company called R2 Technologies, Aylward is helping smooth out one of those bugs with an idea he has for improving mammography. 

His idea is to make it easier for radiologists to detect lesions within every part of the breast on a single image. In traditional mammograms, the breast is compressed, an important step that allows details to show up on the X-ray. But compression isn’t always evenly applied, so if a lesion happens to be in a part of the breast that is not compressed well, then it might be missed by a mammographer. The shape of the breast is also a problem-it makes it difficult for compression to occur evenly. So Aylward uses a technique called “mixture modeling” to compensate for that. 

Mixture modeling involves figuring out where the fat and dense tissue lie within the breast. Then a model is made of each, indicating the difference between the dense tissue and the fat. 

When a mammogram is printed on film, however the breast is compressed is how the film looks. You can’t make any adjustments to it. But now, with computer technology, a mammographer or radiologist can look at the image on the computer screen and actually increase the contrast, emphasizing the detail in a particular area. This is especially useful for examining dense regions of the breast, which usually don’t show much contrast. 

But since radiologists don’t have a lot of time for turning knobs to look at all of the different regions of the breast under a variety of different contrasts, Aylward has designed a program using mixture modeling that automatically adjusts the contrast and brightness within the image. “My program tells the computer how those knobs should be turned,” Aylward says. 

And with that, radiologists should have an easier time detecting breast cancer earlier. “The whole purpose,” Aylward says, “is to raise the bar of the lowest performing radiologist-not replace them, but bring everyone up to a certain level.”

Aylward is pursuing other areas of computer-aided diagnosis. He’s helping Charles Chung, chief of pediatric radiology, diagnose different lung diseases in prematurely born infants. He also collaborates with Guido Gerig, who came to Carolina from Switzerland to accept a joint appointment in psychiatry and computer science. Gerig diagnoses brain diseases based on shapes and structures in the brain. 

But Aylward says his favorite project right now, no question, is the computer-aided diagnosis for mammograpy. “Working with the Department of Radiology and R2 Technologies, my ideas have already paid off,” Aylward says. “Within a couple of months my program could be out in the working world, helping to find cancers earlier, before they really progress.” 

Though Aylward may sometimes consider himself a graduate student-he can still be spotted riding his bike or hanging out on Franklin Street-he's definitely found his niche as a faculty member. He’s even got his own graduate students working for him now, or, as he would prefer to say, “Not working for, but working with.”