Program Information
Monte Carlo Modeling and Simulation of the Varian TrueBeam LINAC Using Heterogeneous Computing
H Lin1*, T Liu1 , L Su2 , C Shi3 , X Tang4 , D Adam5 , B Bednarz5 , X Xu1 (1) Rensselaer Polytechnic Institute, Troy, NY (2) John Hopkins University, Baltimore, MD (3) Memorial Sloan Kettering Cancer Center, Basking Ridge, NJ (4) Memorial Sloan Kettering Cancer Center, West Harrison, NY (5) University of Wisconsin, Madison, WI
Presentations
SU-K-FS1-17 (Sunday, July 30, 2017) 4:00 PM - 6:00 PM Room: Four Seasons 1
Purpose: We herein propose a Monte Carlo-based framework that serves as an accurate and fast dose engine for Intensity-Modulated Radiation Therapy (IMRT). The framework consists of a source modeling kernel that simulates modern medical LINAC and a coupled photon-electron transport kernel originating from our ARCHER Monte Carlo code. The framework is optimized for CPU, Intel MICs and Nvidia GPUs.
Methods: The source model started from phase space files provided by vendors and implemented explicit-approximate transport in secondary jaws and multi-leaf collimators. First Compton scattering process was simulated to account for the contribution from attenuation and the secondary Compton electrons were not tracked. The source modeling kernel was integrated with the Monte Carlo transport kernel asynchronously to improve the performance: source particles were generated for a batch on the fly while the transport kernel was executed for the previous batch. The built source model of TrueBeam was validated against experimental measurements and tested on IMRT plans.
Results: The ARCHER particle transport kernel was verified with homogeneous water phantom and got strictly the same dose results. The validation of TrueBeam source model was performed against measurements and good agreements were found in the percentage depth dose and dose profiles for a range of field sizes. One TrueBeam breast IMRT plan was calculated using ARCHER and DPM. Relative statistical error was kept to be 1% to PTV, and the isodose distributions overlapped well in the whole region. From the aspect of performance, the whole simulation for breast plan takes 49s on a Titan X Pascal GPU and 42s on two K40 GPUs.
Conclusion: We have successfully developed a Monte Carlo-based framework that asynchronously performs the source model and dose calculations for IMRT plans. We are optimizing the framework and dose calculations at various treatment sites are to be presented.
Contact Email: