Core Technology Research For Proton Treatment Planning System

Proton radiotherapy has developed rapidly at home and abroad in recent years due to the Bragg peak characteristics of the depth dose curve of proton beam in tissue.At present,the commercial proton treatment planning system is dominated by foreign companies.Developing domestic made proton treatment planning system with independent intellectual property rights can get rid of the dependence on foreign companies and benefit cancer patients at home.Developing proton treatment planning system faces three major challenges:1)the challenge of finding an optimal scanning path when the number of scanning spots is large;2)the time to complete the dose calculation and dose optimization is long and the efficiency is low;3)the proton pencil beam dose calculation is low when dealing with tumor sites with complex anatomical structure,such as lung and head and neck.Three research tasks were carried to overcome these challenges.In the first research task,we developed a scanning path optimization program based on simulated annealing algorithm,and accelerates the optimization process by using GPU parallel computing technology,and realizes simultaneous scanning path optimization of many energy layers,thus significantly reduce scanning path optimization time.The scan path optimization test results of four clinical cases showed that using the optimized scanning path,the scan path length was reduced by 30.0%,22.6%,55.4% and 62.6%,and the optimization time was 0.217 s,0.249 s,0.173 s and 0.176 s,respectively.In the second research task,dose calculation and dose optimization engine for proton intensity modulated radiotherapy were developed.Based on the pencil beam dose calculation algorithm,a proton dose calculation engine for active scanning irradiation was developed.The pencil beam dose distribution calculated by this dose engine and the Monte Carlo dose distribution calculated by TOPAS had great agreement.We compared the dose distribution of five proton beams with different energies in the homogeneous water phantom.The 2 mm/2% criteria Gamma pass rates between the pencil beam dose distribution calculated by the dose engine and the Monte Carlo dose distribution calculated by the TOPAS were all greater than 98%,while the 3 mm/3% criteria Gamma pass rate all reached 100%.The quasi-newton optimization algorithm was used to find the optimal solution to achieve optimization objectives and accelerates the calculation of function value and scanning spot gradient matrix by using GPU parallel computing techniques.The CUDA C++ programed dose optimization engine was integrated into our domestically made proton treatment planning system.Dose optimization test results of water phantom cases and the clinical case showed that the dose distribution optimized by the optimization engine can achieve optimization objectives.The minimum total time needed for 40 optimization iterations was 30.3 s.In the third research task,we developed a deep learning model that can predict Monte Carlo dose from proton pencil beam dose and CT images,thus rapidly improving the accuracy of proton pencil beam dose calculation to the Monte Carlo level.For one single field,the average prediction time required by the model was less than 3 seconds.The model can accurately predict the Monte Carlo dose for many different types of patients,and can be applied to the new patient data set through simple transfer learning.The Gamma pass rate(1mm/1% criteria)between the predicted dose distribution and the Monte Carlo dose distribution for head neck,liver,lung and prostate test patients reached 92.8%,92.7%,89.7% and 99.6%,respectively.The prediction model can be used as a practical tool to improve the accuracy of proton dose calculation and improve the quality of the treatment plan.In conclusion,this thesis made an in-depth study on the three key technologies of developing proton treatment planning system,including scanning path optimization method,proton dose calculation and dose optimization engine and using artificial intelligence technology to improve the accuracy of proton pencil beam dose calculation,which contributes to the development of our domestic proton treatment planning system.