A sensitivity-guided algorithm for automated determination of IMRT objective function parameters.

Optimizing intensity-modulated radiotherapy (IMRT) plans involves tradeoffs that balance normal-tissue objectives against each other and against tumor objectives. Adjusting the parameters that determine the appropriate contributions of individual anatomic structures to the objective functions through trial and error is time consuming and may not produce the best achievable plans. We have developed a sensitivity-guided parameter optimization (SGPO) method to assist in the automatic determination of parameters to drive the IMRT optimization to better achieve, or even exceed, specified planning goals. The method is based on the trade-off relationships among multiple objectives: In a globally optimal plan (or within a convex subspace of the plan objectives), any attempt to improve the achievement of goals for a structure will result in sacrificing the goals for at least one other structure. However, different objectives may have different sensitivities to the overall goal of an IMRT plan. For instance, changes in dose distribution, hence the subscore corresponding to an objective for a given normal structure, may minimally impact the target dose distribution. Stated differently, the target coverage is insensitive to the changes in dose distribution of the specific normal structure. A lung cancer treatment plan designed with the SGPO method was used to demonstrate that IMRT plans could be designed to favor a structure with the highest target sensitivity and spare the structures with the least target sensitivity without compromising the target coverage. Using one case each of prostate and paranasal sinus cancers, we also demonstrated that several alternative optimal solutions could be designed with the SGPO algorithm favoring different structures. Finally, we applied the method to eight oropharyngeal cancer cases to obtain objective function parameters that satisfied the Radiation Therapy Oncology Group RTOG-H-0022 protocol. The eight plans optimized using the computer-generated objective function parameters met the protocol's scoring criteria with no or only minor protocol violations. Our preliminary study indicates that the SGPO method may be an effective and practical way to improve IMRT planning.