34 Optimization of eMC algorithm parameters for dose calculation in electron

Introduction Monte Carlo algorithms model the transport of particles in matter with great precision considering all physical phenomena which can’t be realized by simple Pencil beam algorithms. Varian Medical System’s eMC algorithm have this stochastic nature but requires significant computing resources. In clinical use, parameters can be adjusted by the user to reduce the calculation time at the cost of dose distribution and monitor unit precision. This study aims to select the group of parameters allowing the best compromise. Methods The combinatorial study of eMC V13.5.35 algorithm parameters examines the accuracy, 1, 2 and 3%, the calculation grid size, 1.0, 1.5, 2.0 and 2.5 mm and the filter smoothing level, Low, Medium and Strong for energies 6, 9, 12 and 16 MeV and fields of 5, 10 and 15 cm 2 , for a total of 432 calculation plans. Percentage depth dose (PDD) and beam profile measured at zref are compared to reference measurements in water by gamma index calculation. Accuracy of monitor unit (UM) number is also evaluated. In addition to the standard fields, ten patient plans with custom cash are studied. Results Local gamma index with 3%/3 mm acceptance criteria of PDD decreases from 98 to 85% for Low to Strong filters only for E6 regardless of accuracy and calculation grid. Analysis with 2%/2 mm acceptance criteria confirms the same trend for E9 with degradation from 88% to 72%. The local gamma index with 3%/3 mm acceptance criteria of beam profiles does not vary by the grid size and accuracy. Analysis with 2%/2 mm acceptance criteria shows a negative effect of the Low filter with 50% of the results below the 95% threshold except for the 10 cm 2 reference field. The differences between MU calculated and measured in a homogeneous water medium increased with Low to Strong filters from 0.8 to 3.0%. Calculation accuracy between 1 and 3% results in increasing differences up to 3.0%. There is no effect of the calculation grid size. Analyzes of calculated MU with custom cache show that the use of an accuracy of 1 or 2% for a 2 mm grid and a Low filter results in MU differences between measured less than 1%. Conclusions The 1%-1 mm-Low parameters provide the most accurate results but have the longest calculation time. The 2%-2 mm-Low parameters are a good compromise on the dose distribution and MU calculated while reducing the computation time by 20.