A GPU-accelerated Monte Carlo dose computation engine for small animal radiotherapy.

BACKGROUND: Accurate dose computation is critical in precision small animal radiotherapy. The Monte Carlo simulation method is the gold standard for radiation dose computation but has not been widely implemented in practice due to its low computation efficiency. PURPOSE: This study aims to develop a GPU-accelerated radiation dose engine (GARDEN) based on the Monte Carlo simulation method for fast and accurate dose computation. METHODS: In the GARDEN simulation, Compton scattering, Rayleigh scattering, and photoelectric effect were considered. The Woodcock tracking algorithm and GPU-specific acceleration techniques were used to obtain a high computational efficiency. Benchmark studies against both Geant4 simulations and experimental measurements were performed for various phantoms and beams. Finally, a conformal arc treatment plan was designed for a lung tumor to further evaluate the accuracy and efficiency in small animal radiotherapy. RESULT: The engine attained a speed-up of 1232 times in a homogeneous water phantom and 935 times in a water-bone-lung heterogeneous phantom when compared with Geant4. Both the depth-dose curves and cross-sectional dose profiles for various radiation field sizes showed a great match between measurements and the GARDEN calculations. For in vivo dose validation, the differences between calculations and measurements in the mouse thorax and abdomen were 2.50% ± 1.50% and 1.56% ± 1.40%, respectively. The computation time for an arc treatment plan delivered from 36 angles was 2 s at a <1% uncertainty level using an NVIDIA GeForce RTX 2060 SUPER GPU. When compared with Geant4, the 3D gamma comparison passing rate was 98.7% at 2%/0.3 mm criteria. CONCLUSION: GARDEN can perform fast and accurate dose computations in heterogeneous tissue environments and is expected to play a vital role in image-guided precision small animal radiotherapy.