Evaluation of a Knowledge-Based Planning Model for Ocular Tumors: Effects of Beam Energy and Mlc Size
Abstract
Purpose
Stereotactic planning for ocular tumors is limited by long optimization times, inter-planner variability, and patient-specific requirements for fractionation, beam energy, and machine configuration. Knowledge-based planning (KBP) can reduce planning time and variability, but changes in fractionation, beam energy, MLC resolution may affect model performance. This study evaluated the robustness of a previously developed multi-fraction ocular KBP model across different beam energies and MLC configurations.
Methods
An in-house RapidPlan ocular model was retrospectively applied to 24 patients. All plans were prescribed 25 Gy in one fraction and normalized to PTV D95% = 25 Gy. For each case, four plans were generated using combinations of 6MV-FFF and 10MV-FFF beams with standard and high-definition HD-MLCs, producing 96 plans total. All plans were created using HyperArc. Plan QA was performed using EPID-based portal dosimetry with a 3%/1 mm criterion. Due to non-normal distributions, results were reported as median, interquartile range, and range. Paired sign tests were used for comparisons, with p < 0.05 considered significant.
Results
HD-MLCs significantly improved plan quality, yielding higher conformity (PCI = 0.89), steeper dose gradients (GI = 3.30), and lower global maximum dose across both beam energies (all p < 0.001). Optic nerve Dmax was reduced with HD-MLCs for both 6MV-FFF (4.77 vs. 3.93 Gy) and 10MV-FFF (5.99 vs. 5.67 Gy) beams (p < 0.001). 10MV-FFF beams significantly improved delivery efficiency, reducing monitor units, modulation factor, and beam-on time by approximately 3 minutes regardless of MLC type (all p < 0.001). Skin Dmax was reduced with 10MV-FFF using standard MLCs (p < 0.001), but not with HD-MLCs. EPID QA pass rates remained high for all plans.
Conclusion
Although trained solely on standard MLC 6MV-FFF ocular plans, the KBP model consistently produced high-quality stereotactic plans and demonstrated improved performance with HD-MLCs and 10MV-FFF beams, supporting its robustness without retraining.