Automatic radiotherapy treatment planning with deep functional reinforcement learning

Background and purpose Intensity-modulated radiation therapy (IMRT) is a crucial radiotherapy technique, often formulated as an optimization problem. However, when the constraints are too tight to provide a feasible solution, human planners resort to relaxing the optimization parameters and re-evaluating until an acceptable solution is obtained. This process is laborious and time-consuming which has prompted attempts to automate radiotherapy through inverse planning studies using reinforcement learning. Unfortunately, these studies face two major limitations. Firstly, a separate sub-network must be designed for each organ, rendering them difficult to apply to patients with an inconsistent number of structures. Secondly, the low signal-to-noise inputs and discrete action space result in low training efficiency. To address these limitations, this study proposes a novel and effective model. Methods This study proposes an organ-sharing network called F unctional a utomatic t reatment P lann I ng N etwork (FatPIN), which contains a functional embedding layer to extract curve features of the dose-volume histogram (DVH). It outputs continuous actions that adjust the optimization parameters, thereby automating the radiotherapy planning process. Results Experiments were conducted on the cervical cancer dataset and the results show that the FatPIN is feasible and effective in real-world radiotherapy. With automatic iteration, FatPIN gradually increased the PTV dose, while reducing the dose levels of the OARs. Specifically, at step 50, the D 95 of the PTV reached 51.68 Gy, exceeding the clinical standard of 50.40 Gy, the V 30 , V 40 and V 50 of all OARs were within clinical requirements. Conclusion We proposed FatPIN to implement automatic radiotherapy treatment planning. Experimental assessments conducted on cases of cervical cancer demonstrate significant improvements in patient metrics facilitated by FatPIN, thus confirming its practical applicability in clinical contexts.