Treatment planning optimization based on response-surface modeling of cost function versus multiple constraints

In treatment planning optimization and inverse planning, a solution is governed by operator-specified objectives and constraints, which are usually in conflict with each other. Currently, the planning process involves manual iterations of testing different combinations of constraints and/or objectives, and evaluating the resulting solutions. In this work, the authors are developing a more efficient paradigm to automate this process. The key component of their approach is to model the cost to be optimized as a function of all the competing constraints. Initially the authors evaluated this technique for optimizing angles and weights of coplanar beams in conformal therapy of pancreas and prostrate cancer. The minimum tumor dose was maximized with multiple dose-volume constraints for other critical structures. An initial set of constraints was established such that the tumor dose fell within desired limits. Subsequently, each active constraint was changed by a small step and the optimization was rerun to obtain a new solution. Effectively, the authors sampled the solution space and obtained a database of the optimized solutions for a variety of different constraint values. They then modeled the tumor dose as a multi-variable function of doses to critical structures using a response-surface method. This allowed the authors to visualize how the constraints competed and affected the tumor dose. Using this information, one can interactively navigate through the solution space and quickly balance among the tumor dose and doses to other structures. This new paradigm focuses on clinical considerations during optimization and improves the efficiency of the treatment planning process