Development of fast Monte Carlo code for high dose rate brachytherapy

Brachytherapy plays an important role in the treatment of various cancers. The TG-43 formalism is widely used in dose calculation for brachytherapy, but it is not accurate as it assumes the patient as homogeneously constituted of water. Monte Carlo simulation can obtain accurate result, but computing time using traditional code is too long to be accepted in clinical application. This work developed a fast Monte Carlo code THUBrachy for high dose rate brachytherapy dose calculation. THUBrachy can simulate photon with energy up to 3MeV for heterogeneous materials and obtain dose, fluence and dose distribution. It applies several accelerators such as Graphics Processing Unit (GPU), Intel Xeon Phi to perform dose calculation with good accuracy and high speed. THUBrachy adopts several parallel programming models to transplant the code onto different hardware accelerators. Voxel geometry based on CT images is used to describe the tissues of patient. A linear track-length estimator is used for dose calculation. In the preliminary accuracy test, a cuboid water model and a heterogeneous-material model were selected as the phantoms. Dose and fluence distributions were calculated by both THUBrachy and FLUKA which is a well-benchmarked Monte Carlo code. A total of 5x108 photons were simulated to make the uncertainty in all area less than 5%. The maxium difference of the results in low dose area is less than 3% and in high dose area is less than 1%, indicating a good agreement. In the performance test, a phantom consisted by 24 types of human tissue materials was chosen to estimate the performance in practical application. Dose distributions with an uncertainty of 2% or less for the target volume were obtained in less than 1 min by simulating 5x107 photons with a GTX 1080Ti, while FLUKA taking several hours to simulate the same number of particles. There is great potential for clinical application in brachytherapy dose planning using this fast Monte Carlo code.