LAUSR

Introduction In the framework of VMAT clinical plan verification with PRIMO, gamma analysis is one of the main tool to check correspondence of dose distributions. Statistical noise (inherent to Monte Carlo dose calculations) influences gamma index. Gamma index is underestimated when reference dose contains noise. It is overestimated when evaluated dose contains noise. Methods We compared dose calculated with PRIMO 0.3.1 (DPM) against Eclipse (Acuros 13.7, dose to medium). Latest PRIMO beta versions include a “macro” mode. It allows from a simple command file, to easily perform all a actions needed to start a simulation, including setting calculation parameter like number of histories, variance reduction (splitting) and parameter to convert to absolute dose. This macro mode allows to calculate gamma index with a given dose distribution, including activating a noise reduction filter (IRON) integrated in PRIMO. We calculated gamma index (for 1 clinical case) for 3 situations: Acuros as reference, PRIMO as reference, with and without noise reduction filter. We vary the statistical noise by either changing the number of histories (from 0.5 × 10 9 to 9 × 10 9 ) and the splitting parameter (from 200 to 400). We calculated gamma index with different parameters: 3%/3 mm, 2%/2 mm and 1%/2 mm. Results Statistical uncertainty varies from 8.4 to 1.7%. Gamma index is systematically higher with Acuros as reference. Applying noise reduction filter systematically and significantly improve gamma index. With 3 ways to reduce noise (and so to minimize its influence on gamma index), the cheapest (regarding calculation time) is the noise reduction filter, and then splitting (with a saturation effect with a value around a few hundreds), and finally an increasing histories number. Noise reduction filter has the same effect than an increase with a 5 to 10 factor of simulated histories. Conclusions In order to compare PRIMO and Acuros dose, the noise reduction filter integrated to PRIMO (IRON) is an efficient tool to minimize influence of statistical noise on gamma index analysis. A detailed analysis of data combined with generalization to other clinical cases will lead to the most efficient balancing between splitting and number of histories to reduce statistical noise.