Before he was a physician, Julian Rosenman was a physicist. It’s no wonder, then, that as he races around checking patient x rays and answering questions about treatment options, his mind is also racing to figure out ways to use the basic research of physicists to help cancer patients.

Rosenman, a professor of radiation oncology who specializes in lung cancer, explains that the three major ways to treat lung cancer are surgery, removing part or all of one lung; chemotherapy, using drugs to kill cancer cells; and radiation therapy, using high-energy beams to kill cancer cells and shrink tumors. Usually, doctors use a combination of the three.

As a radiation oncologist, Rosenman is interested in making radiation doses more accurate and faster to calculate. The idea behind radiation therapy is simple: find the tumor and aim a bunch of radiation beams at it from different angles. But the difficulty, especially with the lung and all of its air pockets, is making sure that the tumor gets a high dose of radiation without causing damage to the surrounding tissue.

To make targeting tumors easier, Carolina’s computer science department helped develop a software program that models a patient in three dimensions, so technicians can find the exact placement of the tumor. Called 3D-conformal therapy, the software came out in the 1980s and is being used not only at Carolina but also at many other medical institutions to help treat prostate and lung cancers.

More recently, radiation oncologists have been able to take radiation beams and break them up into hundreds of little beamlets. This technique, known as intensity modulation, allows the oncologists to assign different weights to each beamlet, so they can add more strength to specific areas. “The clinical value of this is just beginning to trickle in,” Rosenman says. “And it’s showing that you can give higher radiation doses with fewer complications, and that’s good.”

While these techniques are proving helpful, the methods used to calculate radiation doses are still very approximate. There is a way to do the calculations correctly, Rosenman explains, but even with today’s computers, the calculations are so complex that they can’t be counted accurately in a reasonable amount of time. “One calculation could run for months on a computer, and the patient could die in that amount of time,” he says.

The method used to correctly calculate doses is actually the same method that’s been used for decades by the armed forces to predict what will happen when an atomic bomb is dropped—for example, how far and in which direction the rays will spread. It goes by the term Monte Carlo and can be used in many situations, not just for radiation and physics, but for things such as stock market analysis and flow control. “But the army isn’t treating patients,” Rosenman says. “So it’s not as urgent for them to do calculations as quickly.”

In terms of radiation, when a tiny particle of light (a photon) from an x-ray machine enters a person’s body, many things can happen: it can scatter, it can be absorbed, it can produce an electron. And one photon may do up to 1,000 different things. This is where Monte Carlo comes in: it calculates the probability for each of those things to happen to each photon as it enters the body and then puts the information into a table. “After awhile you start to get a pattern of what is really likely to happen,” Rosenman says.

Knowing there was a reliable, though slow, method out there for generating accurate dose calculations, Rosenman’s physics-oriented mind wasn’t going to rest until he could figure out a way to make the method fast enough to treat patients. His idea: break up the calculations and run them on lots of computers at one time.

But how many computers would it take and how was he going to pay for them? Fortunately, about the same time Rosenman came up with his idea, Microsoft and Dell happened to be on campus looking for interesting projects. So Rosenman pitched his idea, and they asked him on the spot how many computers he thought it would take. He asked for 100. “I asked for more than I thought I would need,” Rosenman says. “But it turns out that one hundred is exactly the number we needed. Fifty would have been too few and one hundred fifty would have taken up too much space. It was a lucky guess.”

A benefit for Microsoft is that the code will be run on a Windows NT system, which will help Microsoft solve some of its own software problems, such as how to program the software to inform another machine that some of the machines have crashed. “Our common interest is trying to make one hundred machines behave as a single unit,” Rosenman says.

Now that they have the computers, Rosenman and his team are working on how to divide the calculations so that they can be run on different machines all at the same time and then added together at the end. The idea, explains Eric Schreiber, a nuclear physicist directing the project, is to have some of the computers running calculations for a beamlet coming in from one direction while other computers are working on a beamlet coming in from another direction.

The researchers are hoping the Monte Carlo code will also enable them to treat patients with electron beams, which work differently than x rays in that they stop once they’ve penetrated the tumor instead of going all the way through, which means less damage to the tissue behind the tumor. But calculating doses for electron beams is more difficult than for x rays. Even the approximation methods that have been used for x rays are not satisfactory. “So nobody uses them right now,” Rosenman says. “We’re hoping Monte Carlo can help with that.”

The goal is to get the computers working together and in the clinic within a year. “But before we can treat patients,” Schreiber says, “we have to be very careful that the results we’re getting are as close to the exact model of the real human being as we can get. If we’ve made part of the model even one sixteenth of an inch too thick, or if there’s a small glitch in the code, then that would affect the outcome tremendously.”