Research On Automatic Intensity-Modulated Radiotherapy Algorithm Based On Reinforcement Learning

Radiotherapy planning is the most critical step in the radiotherapy process.The physicists repeatedly try the planning parameters in the dose optimization problem,such as the weight,volume and dose value of the target and the organs at risk.The goal is to achieve the clinical requirements of the dose of the tumor and reduce the dose to the normal tissue.Due to the problems of time-consuming and inconsistent plan quality in traditional manual planning,there have been studies on the feasibility of automatic IMRT planning technology based on statistical or deep learning methods to predict plan parameters.However,existing methods have some problems,such as low automation,low training efficiency,and insufficient application of models.Therefore,an Automatic Treatment Parameters Adjustment Network(ATPAN)based on reinforcement learning is proposed.The algorithm builds a virtual radiotherapy planning environment based on the real plan design scenario,and used the interaction between agents and environment to obtain the data series required for model training.In addition,combined with the functional linear regression model,we takes the DVH curves as the model input after the functional feature transformation,so that the curve features are fully extracted.The model provides guidance for parameter adjustment according to the learned behavior value strategy.Finally,we implements automatic planning under Deep Q-Network(DQN)and Actor-Critic(AC)reinforcement learning frameworks respectively.The feasibility and effectiveness of the model are verified from test data,and the influence of different reward functions on model is analyzed.From the experimental results,compared with before parameter adjustment,ATPAN can significantly improve the dose distribution and dosimetric parameters of patients.Therefore,the method can not only realize the automatic adjustment of radiotherapy planning parameters,but also for practical application,ATPAN can reduce the labor time,narrow the difference between the quality of plans,and improve the quality of radiotherapy plans.