Physicians use parallel processors in radiotherapy analysis
Physicians use parallel processors in radiotherapy analysis
During the past few years, 700,000 Europeans have died of cancer, and more than 1.2 million have contracted the disease. About 25% of the new patients each year have tried radiosurgery as a cure, but 20% of these treatments failed because the tumors were not precisely located or the irradiation missed complete coverage.
To help reduce these failures, a European consortium, known as the Radiotherapy Application Deployment on Parallel Technology (RAPT), is exploring the use of more powerful parallel processors in radiosurgery simulation systems. These systems will enable clinicians to test their proposed treatment plans on computer models of patients before actual use.
The result of a joint development between Electronic Data Systems (EDS; Italy), the UK Parallel Applications Center (PAC), and Parsytec (Germany), a new parallel-processor system can simulate how complex arrays of beams are affected by their passage through different types of human tissue. It also informs radiotherapists whether sufficient energy has reached every part of the tumor for complete destruction and how much damage has been done to nearby healthy tissue. The radiotherapist is then able to see the results of each test as a three-dimensional image that can be manipulated in real time.
In operation, the system captures images from computer-tomographic scans and stores this information with other patient data for graphical display. A three-dimensional image representing the patient`s head, for example, can then be constructed to provide the necessary input data for a simulation process (see figure on p. 12).
To calculate the predicted radiation dosage, Monte Carlo simulation codes, originally developed in large particle physics research centers, are used. At present, such simulations for patient treatment are feasible only by supercomputers, restricting the technique to just a few cases at large hospitals.
The radiotherapist then provides a description of the treatment session, including type of accelerator, number of beams, beam orientation and intensity, and patient anatomy. The simulation system computes the absorbed dosage at each point in the patient`s anatomy using a beam characterization for the specified accelerator.
Visualization software allows the inspection of the absorbed dose values calculated by the simulation. The radiotherapist has the freedom of choosing various visualization modes, which permit the inspection of the results from different perspectives and using different techniques.
"By using a cluster of Digital Equipment Corp. (Maynard, MA) Alpha workstations, built from inexpensive machines and interconnected by a Fiber Distributed-Data Interface channel," says Mike Surridge of PAC, "the RAPT application code can be ported to other workstations, such as those from Sun Microsystems, IBM, and Hewlett-Packard. Our main problem was how to fit the anatomical dataset into the smallest possible memory, because memory cannot be expanded much on low-end processors," he adds.
High simulation accuracy implies that the Monte Carlo calculation has to compute a large number of photon histories in the selected media. The RAPT code works by dividing the photons among a large number of inexpensive computers and combining the results from each computer at the end of the process. Each computer handles a fraction of the photons, so that such simulations can be carried out with greater accuracy using millions of photons in total, in a run-time of around one hour.
"This will make it practical to carry out full Monte Carlo analysis of radiotherapy treatments for individual patients on a machine that will be affordable for a typical, medium-sized medical physics department," says Surridge.