Iterative prescription refinement in fully discretized inverse problems of radiation therapy planning

We formulate the method of iterative prescription refinement for inverse planning in any fully discretized model of radiation therapy. The method starts out from an ideal dose prescription and repeatedly refines it into a refined dose prescription. This is done computationally without human interaction until a prespecified stopping rule is met, at which point the refined dose vector and the accompanying beamlet intensities vector are evaluated and presented to the planner. The algorithmic regime is general enough to encompass various physical models that may use different particles (photons, protons, etc.) It is formulated for a general inversion operator thus different objective functions or approaches to the optimization problem (such as DVH, gEUD, or TCP and NTCP cost functions) may all be applied. Although not limited to this model, we demonstrate that the approach at all works on two exemplary cases from photon intensity-modulated radiation therapy.