by Roy Keyes
Cancer is a devastating group of diseases that affects nearly everyone, whether directly or indirectly. In the developed world, cancer is the number three cause of death. Fortunately, in the last thirty-five years cancer mortality rates have fallen dramatically, due to improvements in both preventive measures and therapies. One of the most effective and widely used treatments is radiation therapy. Radiation therapy technologies have made rapid improvements, largely due to advances in computing. I intended to use AMD's newest multi-core OpteronTM processors to bring radiation cancer treatments to a new level of accuracy and curative power.
The goal of radiation therapy is to deliver a lethal dose of radiation to the tumor while sparing healthy tissue as much as possible. Ideally, a radiation therapy dose distribution would put the exact desired dose inside the tumor and nowhere else (see figure 1), but physics places fundamental restrictions on what is achievable. In order to deliver the best possible dose distribution, high resolution imaging, such as CT or MRI, is used to map out the target geometry and composition. After the radiation type and delivery technique is chosen, we must understand how the radiation will interact with the patient anatomy, so that optimal beam parameters can be chosen.
Figure 1: The ideal radiation dose distribution for radiation therapy.
Radiation interactions with human tissue are usually approximated by algorithms that assume humans are composed entirely of water. While this works relatively well in many cases, it does a very poor job of dealing with tissues that are not very water-like, such as the lungs, bone, or with metal implants. These difficulties can be overcome by using Monte Carlo algorithms to simulate the interaction of radiation with tissue on the atomic level. Although capable of producing extremely accurate results, Monte Carlo algorithms are almost never used clinically for one reason: they are very slow.
Figure 2: A Monte Carlo based proton beam treatment plan using a simulated patient.
Fortunately, accurate Monte Carlo dose calculations do not have to be slow. Monte Carlo calculations have the great property that they can be easily scaled with parallel processors. This means that by taking advantage of a large multi-core architecture, such as AMD's new OpteronTM 6100, clinically usable calculation times can be achieved.
As part of my PhD research at the University of New Mexico, I am working on algorithms for calculating radiation dose distributions that will make use of dedicated, massive multi-core machines and clusters. Massive multi-core machines provide the most direct route to achieving time scales useful to cancer clinics for the most advanced radiation dose calculations, bypassing both networking issues inherent in clusters and unavoidable programming hurdles with GPU's. With 48 AMD cores, I plan to demonstrate that cancer clinics can finally produce the most accurate radiation therapy dose calculations, which will allow patients to receive more accurate and effective cancer treatments with fewer adverse effects.