A Fast Monte Carlo Dose Algorithm for Radiotherapy Treatment Planning Based On Hybrid Adaptive Meshes
J Yuan*, J Brindle, Y Zheng, J Sohn, P Geis, M Yao, S Lo, B Wessels, University Hospitals, Case Medical Center, Case Western University, Cleveland, OHSU-C-211-2 Sunday 1:30:00 PM - 2:15:00 PM Room: 211
Purpose: Monte Carlo methods are considered to be the most accurate dose algorithm for radiotherapy. Variance reduction techniques such as history repetition, Russian roulette and photon splitting are employed to improve the calculation efficiency. Generally, it takes a large portion of the simulation time for two inevitable tasks, that is, voxel-to-voxel boundary crossing and energy deposition. The purpose of this work is to investigate the potential for additional speedup achieved by reducing the number of boundary crossing based on hybrid adaptive meshes. Method and Materials: A Monte Carlo code was developed to simulate the coupled photon-electron transport for radiation therapy using a hybrid adaptive mesh. Photon transport was modeled in an analog fashion. The Compton scattering, photoelectric ionization and pair production were considered. For electron transport, a condensed history method was used in which the hard interactions such as inelastic collision and bremsstrahlung were simulated explicitly. The formulation by Kawrakow and Bielajew was used for electron multiple scattering. Photons and electrons were tracked on a hybrid adaptive mesh, which was generated from an initial uniform coarse mesh. The coarse uniform mesh was divided voxel-by-voxel into an unstructured finer mesh depending on the splitting criterion such as the density gradient. The resulting adaptive mesh contains larger voxels in smooth density areas and smaller voxels used for density regions with large gradient to retain the accuracy. Results: The speed up observed by varying the splitting level is proportional to N1/3, where N is the total number of the voxels. For the test cases, 30% of the calculation time was saved by using the adaptive meshes starting with 10mm spacing and reduced to 1.25mm voxels for high gradient regions comparing with the 1.25mm uniform meshes. Conclusions: The Monte Carlo simulations can be further accelerated based on these hybrid adaptive meshes.