Computer-assisted beam modeling for particle therapy.

Abstract

PurposeTo develop a computer‐driven and thus less user‐dependent method, allowing for a simple and straightforward generation of a Monte Carlo (MC) beam model of a scanned proton and carbon ion beam delivery system.MethodsIn a first step, experimental measurements were performed for proton and carbon ion energies in the available energy ranges. Data included depth dose profiles measured in water and spot sizes in air at various isocenter distances. Using an automated regularization‐based optimization process (AUTO‐BEAM), GATE/Geant4 beam models of the respective beam lines were generated. These were obtained sequentially by using least square weighting functions with and without regularization, to iteratively tune the beam parameters energy, energy spread, beam sigma, divergence, and emittance until a user‐defined agreement was reached. Based on the parameter tuning for a set of energies, a beam model was semi‐automatically generated. The resulting beam models were validated for all centers comparing to independent measurements of laterally integrated depth dose curves and spot sizes in air. For one representative center, three‐dimensional dose cubes were measured and compared to simulations. The method was applied on one research as well as four different clinical beam lines for proton and carbon ions of three different particle therapy centers using synchrotron or cyclotron accelerator systems: (a) MedAustron ion therapy center, (b) University Proton Therapy Dresden, and (c) Center Antoine Lacassagne Nice.ResultsParticle beam ranges in the MC beam models agreed on average within 0.2 mm compared to measurements for all energies and beam lines. Spot sizes in air (full‐width at half maximum) at all positions differed by less than 0.4% from the measurements. Dose calculation with the beam model for the clinical beam line at MedAustron agreed better than 1.7% in absolute dose for a representative clinical case treated with protons. For protons, beam model generation, including geometry creation, data conversion, and validation, was possible within three working days. The number of iterations required for the optimization process to converge, was found to be similar for all beam line geometries and particle types.ConclusionThe presented method was demonstrated to work independently of the beam optics behavior of the different beam lines, particle types, and geometries. Furthermore, it is suitable for non‐expert users and requires only limited user interaction. Beam model validation for different beam lines based on different beam delivery systems, showed good agreement.