gms | German Medical Science

ISTC Project #3563 aims to develop a Monte Carlo solver to be incorporated into a Treatment Planning System (TPS) used at the proton therapy center operated at the Institute for Theoretical and Experimental Physics (ITEP), Moscow. The speed of computing is an important criterion of modern numerical methods used in TPS because it is desirable to minimize the time needed to prepare patients to radiation treatment. This becomes extremely important in complicated cases. The higher accuracy of calculations increases the time, especially in the real-time mode. One of the alternatives to resolve the problem is to develop specialized solvers.

The paper describes the objectives of and directions in the development of a specialized processor for Monte Carlo calculations. It also describes why the approach we have chosen is better than the development of a PC cluster. The solver is based on a large FPGA (Altera, Stratix3), hierarchic memory to optimize access to data (DDR SDRAM, FPGA internal memory), built-in applications (PCI interface), hardware treatment of the entire problem, and a call library from the end user’s code for ease of use.

Specific features in the architecture of the specialized Monet Carlo solver for proton therapy are considered. A flow chart of the large-FPGA-based solver is provided.

Theoretical evaluation is made on possible acceleration for the class of problems under consideration in comparison with PC implementations. Accelerations obtained in the simulation of some fragments of the computational algorithm are reported. Acceleration is reached due to the parallel implementation of basic operations (generation of new particle coordinates, determination of points where particles cross cell boundaries, collection of statistics etc.) in the specialized modules of the conveyor solver.

Provided are results for two test calculations: random number generation and particle crossing of a plane.

The work was in part supported by the ISTC, Grant #3563.