A Correction of the Gamma Test for Monte Carlo Dose Distributions Comparison

PURPOSE For QA and commissioning purposes, the gamma index is often used to compare two dose distributions. However, this method is known to be asymetric and dependent on the resolution and noise. This work proposes a correction to unbias the gamma test applied to noisy Monte Carlo (MC) doses, while keeping a reasonable computation time. MATERIALS AND METHODS The noise dependency of the gamma index is first analyzed by computing the gamma passing rate for multiple MC doses with various noise levels. These results are then fitted with a function derived theoretically. This fit enables to extrapolate the gamma index for a low noise case, which would normally require to simulate a large number of particles with MC. To illustrate the method, four proton treatment doses (brain, prostate, lung, liver) were recalculated with the fast Monte Carlo code MCsquare and compared to the TPS doses using gamma tests (2%/2mm, 3%/3mm, 4%/4mm, with MC as reference). The noiseless gamma passing rates were then estimated using the proposed method, by fitting gamma values obtained for 1E7, 2E7, ...[À compléter] simulated particles. This prediction is then validated with the passing rates obtained for low noise MC doses (1E9 particles). RESULTS On average, the difference betwen gamma passing rates obtained for 2E7 and for 1E9 particles simulations was ...[À compléter]. The MC computation times were around 2 minutes and 2 hours respectively. For all patients, the absolute error between the real and the predicted passing rates remained between 3% and was around 1% in most cases. CONCLUSIONS This work offers an alternative method to the classical gamma index when there is noise in the reference dose. It shows good results on four proton treatment plans with, in some cases, quite important gains of precision while keeping the computation time practicable