Abstract
IMRT treatment planning via biological objectives gives rise to constrained nonlinear optimization problems. We consider formulations with nonlinear objectives based on the equivalent uniform dose (EUD), with bound constraints on the beamlet weights, and describe fast, flexible variants of the two-metric gradient-projection approach for solving them efficiently and in a mathematically sound manner. We conclude that an approach that calculates the Newton component of the step iteratively, by means of the conjugate-gradient algorithm and an implicit representation of the Hessian matrix, is most effective. We also present an efficient heuristic for obtaining an approximate solution with a smoother distribution of beamlet weights. The effectiveness of the methods is verified by testing on a medium-scale clinical case.
Export citation and abstract BibTeX RIS