Evaluation of the variance reduction options available with the new beam code distribution

Full text: The BEAM Monte Carlo code has been widely benchmarked for dosimetry calculations with medical linear accelerators. The computational time required to run simulations on BEAM remains a limitation to its use in the clinic. The computational time can be reduced by increasing a) computing power and/or b) calculation efficiency. Computing power may be increased by the use of PC clusters which require significant resources and expertise that may not be available in most small radiotherapy clinics. Calculation efficiency is improved by variance reduction and approximation techniques. The latest distribution, BEAMnrc04 contains new variance reduction options that are claimed to improve photon fluence efficiency by a factor of 8 over the previous distribution. This work investigates the efficiency gains available with BEAMnrc04 for simulations of megavoltage photon beams. A model of a high energy (16 MV) and a low energy (6 MV) linear accelerator were investigated. The 16 MV model is a simplistic example provided by the BEAM developers. The 6MV model was created for this work from Varian's specifications of a Clinac 600C. The efficiency was calculated for each case using the following equation: ε = 1/(Ts2). Where ε is the efficiency, T is computer processing time and s is the estimated uncertainty in fluence in a 2 cm square scoring region on the central axis at 100 cm distance. The uncertainty was calculated by the BEAM code system. The new bremsstrahlung splitting technique called Directional Bremsstrahlung Splitting (DBS) is compared for efficiency with the older Uniform Bremsstrahlung Splitting (UBS) method. Photon Forcing, Russian Roulette (RR) and Range Rejection were also tested for their impact on efficiency. The same simulations were run on two desktop computers, one with a 1.7 GHz processor and one with a dual processor (2 x 3.06 GHz) for comparison. All efficiencies are quoted relative to a reference simulation with default settings and no bremsstrahlung splitting. Changing Range Rejection (Global electron cut-off energy) from 1 MeV to 3 MeV increased photon efficiency by a factor of 1.41. No effect on electron efficiency was observed. Photon forcing made no significant difference in photon fluence efficiency with photon forcing turned on. A decrease in electron fluence uncertainty was observed but not enough to significantly improve electron efficiency when using DBS with esplitting. Maximum photon efficiency with DBS (no e-splitting) was 13.5 times greater than with UBS (no Russian Roulette) and 176 times greater than with no bremsstrahlung splitting at all. With e-splitting and RR enabled the gain with DBS over UBS was reduced to 8.7 times. No significant difference in electron efficiency was observed. The effect of e-splitting when using DBS is to decrease photon efficiency by a factor of about 0.9. No significant difference in electron efficiency was observed. The effect of moving the RR plain relative to the splitting plane with DBS was not significant. The largest efficiency was achieved with the splitting plain on the bottom surface of the flattening filter and the RR plane 1mm above it. The dual processor computer was on average a factor of 2 times more efficient than the 1.7 GHz computer. Efficiency gains of about two orders of magnitude can be achieved with the use of DBS. A gain of approximately 9 times was achieved with the DBS technique over UBS. Continued improvement of variance reduction techniques with BEAM will significantly improve the efficiency of phase space and ultimately dose calculations. Copyright (2004) Australasian College of Physical Scientists and Engineers in Medicine