Knowledge-Based Stereotactic Planning for Single-Isocenter Multiple Brain Metastases

Knowledge-based planning (KBP) approaches have been shown to reduce variability in plan quality by providing estimates of dose-volume histograms (DVHs) and inverse planning objectives. We established and validated a KBP model using KPB software for multiple brain metastases. Forty-nine previously planned patient cases, each containing up to four brain metastases, were selected at random and used to train a KPB software model for estimation of DVHs of brainstem, whole brain, optic chiasm, optic nerves, and eyes. All training cases had been planned using volumetric modulated arc therapy (VMAT) with a single isocenter. The model was validated by replanning 16 independent cases, in which line objectives for optimization were derived from estimated DVHs. Acceptable KBP-informed treatment plans were obtained for 14 of 16 cases without interaction by the treatment planner during optimization. Organs at risk (OARs) showed lower maximum doses compared to previously generated clinical plans, in some cases by over 4 Gy; however, differences were insignificant on average (P>0.5). Model-estimated DVH bands contained, or were more demanding, than those for clinical cases for 60% of OARs. When line objectives based on the lower bounds of estimated DVH bands were used to drive optimization, the estimate was met or surpassed for 70% of OARs. Planning target volume coverage was slightly worse for KBP compared to clinical plans, however this difference was also insignificant (P>0.1) on average. Cases for which the model required intervention by the planner included those with a larger number (>4) of metastases than used in training, and for which a target volume was immediately proximal to an OAR. Knowledge-based treatment planning with KPB software is capable of producing realistic DVH estimates and optimization objectives that yield clinically acceptable treatment plans with minimal user interaction. Further model training, including cases with more metastases and those proximal to OARs, may improve versatility.