Fully Automated Multicriteria Optimization (MCO) Treatment Plan Generation for Radiation Treatment Planning

Abstract

To develop a novel multi-criteria optimization (MCO) solution approach to generate treatment plans without Pareto front navigation (any user-computer interaction). Current MCO algorithms consider convex optimization problems and create an approximation of the Pareto front and then navigate over the approximation to obtain the final treatment plan. Our proposed method is able to handle nonconvex optimization optimization problems which includes dose-volume objectives/constraints, tumor control probability (TCP), normal tissue complication probability (NTCP), etc. In addition, instead of creating an approximation and navigation, our method targets the solution without any time-consuming human-computer interaction steps. This is achieved by a solution approach that utilizes given dose bounds on the structures by clinicians. We tested our method with 10 locally advanced head-and-neck cancer patients retrospectively using the concept of segment weight optimization. Monitor units were reoptimized using the segments extracted from the conventional intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT). We compared the results with conventional clinical plans and current MCO approach. Our MCO formulation had total 9 objective functions, which included 3 distinct objectives for primary target volume, high-risk and low-risk target volumes, 5 objectives for each of the organs-at-risk (OAR) (two parotid glands, spinal cord, brain stem, and oral cavity), and one for the non-target non-OAR normal tissue. Compared to the conventional clinical plan and current MCO method, the proposed MCO method achieved average reductions in left parotid mean dose of 8% (P < 0.01) and 7% (P < 0.01), right parotid mean dose of 12% (P < 0.01), and 8% (P = 0.02), oral cavity mean dose of 1% (P = 0.63) and 5% (P = 0.28) and spinal cord maximum dose of 28% (P < 0.01) and 10% (P = 0.02), brain stem maximum dose of 41% (P < 0.01) and 18% (P = 0.01), and normal tissue maximum dose of 9% (P < 0.01) and 8% (P < 0.01), respectively. We proposed a novel MCO solution approach that generated superior high quality IMRT and VMAT treatment plans without any human-computer interactions.