Room: Exhibit Hall
Purpose: Monte Carlo (MC) based calculations are considered the gold standard for dose calculation accuracy in radiation therapy. Parallelization of the MC codes with available source codes is a common approach to increase performance. We proposed to optimize the efficacy of simulation with the serial codes without available source codes by splitting the spot scanning sequence for a field.
Methods: Patient treatment plans were created in Eclipse treatment planning system (TPS) (Varian Medical Systems Inc. Palo Alto, CA, USA). The parameter files for the TOol for PArticle Simulation (TOPAS) Monte Carlo code were generated with an in-house MATLAB script from the exported DICOM RT file. Spot scanning sequence for each field was split into 2, 4, 8, 16, and 32 sub-beams. Parameter files were dispatched at-shared High Performance Computing cluster with IBM Spectrum LSF Standard 10.1.0.0 software. The data on time and memory usage were acquired from TOPAS log files.
Results: The median calculation times were 6.73 and 18.4 hours for phase space calculation and dose and linear energy transfer (LET) scoring in patient CT images, respectively, for the serial version of the code with the un-split parameter file. Median times for phase space calculations were 1.76, 092, 0.46 and 0.22 hours and dose/LET scoring times were 4.6, 2.7, 1.4, and 0.7 for sub-beam splits of 4, 8, 16 and 32, respectively, for a 4 field test case. The proposed method assured speed up of 20 times compared to un-split calculations thus optimizing the computations of the treatment plan verification with the serial general purpose Monte Carlo codes.
Conclusion: Our approach was able to reduce patient Monte Carlo dose calculation times. Given the time savings, dose and LET scoring may be utilized to better understand clinical treatment plans without disrupting the current planning workflow.