Automatic Planning Of Intensity-Modulated Radiotherapy Based On Optimization Parameters Search Using Artificial Intelligence

Intensity-modulated radiation therapy(IMRT)based on reverse planning has become the major treatment technique of radiotherapy.Medical physicists adjust multiple parameters,e.g.the number of the beams,beam angles,and dose constraints on targets and organs at risk,etc.,to determine the objective function and an optimization is performed to minimize/maximize the objective function.In a trial and error manner,the clinical requirements are met,and the physicists complete the treatment plan design through a forward dose calculation.The overall purpose is to reduce the dose of the surrounding normal tissue as much as possible while delivering lethal dose to the tumor.In the whole planning process,optimization remains as the core phase,where the physicists need to spend a lot of time finding a good set of optimization objective parameters.Depending on the individual’s clinical experience and skills,the quality of the plans from different physicists may vary greatly.Therefore,the automatic planning has played an important role in simplifying optimization process,reducing the differences between individual and inter-hospital,improving the plan quality and efficiency,and realizing homogenized radiotherapy.To solve the above problems,this work uses VARIAN’s Eclipse(Varian Medical Systems,Palo Alto,CA,US)Treatment Planning System(TPS)and its Eclipse Scripting Application Programming Interface(ESAPI)to realize data access.This work is based on two different ideas to carry out research on IMRT automatic planning,and the specific contents are as follows:1、In this study,the Optimization Parameter Tree Search Algorithm(OPTSA)was proposed,which simulated the trial-and-error process of physicists in optimization by iterations in order to find the optimal optimization objective parameter set.With the help of ESAPI scripting tool,OPTSA could automatically complete all steps,including planning creation,parameters setting,planning optimization and dose calculation.Finally,a complete IMRT automatic planning was generated.The results have showed that the OPTSA’s automatic planning could meet the clinical requirements for rectal cancer and cervical cancer.The quality of automatic planning could reach the same level as manual planning,or even better.Compared to manual planning,automatic planning based on OPTSA has a unique advantage in terms of saving labor cost.2、Based on the framework of deep reinforcement learning(DRL),the Optimization Adjustment Policy Network(OAPN)was proposed to automate the process of planning optimization.OAPN used ESAPI to realize data interaction with TPS,and used clinical data to carry out network training.According to the learned action-value function,OAPN could efficiently adjust the optimization objective parameters and realize the automation of high-quality planning.The results have shown that the strategy of adjusting optimization objective parameters learned by OAPN was similar to the idea of manual planning,and it had the potential to improve the plan quality.In addition,the average time of planning design was about 4 minutes for all test cases,which greatly improved clinical efficiency.Modularized program implementation of the aforementioned idea was realized by ESAPI in this work.While realizing the IMRT automatic planning,it can effectively reduce the labor time,narrow the differences between individual and inter-hospital,improve the radiotherapy plan quality,and is expected to improve the efficacy and quality of life for cancer patients.