Monte Carlo simulations on GPU for brachytherapy applications

In brachytherapy, dosimetric plans are routinely calculated with the TG43 formalism which considers the patient as a simple water box. However, accurate modelling of the physical processes considering patient heterogeneity using Monte Carlo (MC) methods is currently too time-consuming and computationally demanding to be routinely used. As a solution we implemented an accurate and fast MC simulation based on Geant4 on graphics processing units (GPU) for brachytherapy applications. Existing approaches using GPU architecture for brachytherapy MC simulations suffer from numerous approximations including, the use of virtual seed bound to a phase space file to define dwell sources, or removing voxel within the CT image to include seed density. Within the proposed framework such approximations have been removed. A comparison between dosimetric plans based on the current clinical standard (TG43) and the proposed full MC simulation led to substantial differences due to the TG43 related approximations assuming the patient as a water box. Finally, the proposed dosimetry platform is capable of providing accurate dose distributions within one minute, which is compatible for a clinical routine usage.