Multifractionation knowledge-based planning model for the stereotactic treatment of ocular tumors

Linear accelerator-based stereotactic radiosurgery and stereotactic radiotherapy (SRS/SRT) are commonly used in the treatment of ocular malignancies. However, a substantial portion of the treatment planning process relies on manual plan optimization to achieve acceptable plan quality, leading to increased planning time and variability among planners. In this work, we demonstrate the feasibility of adapting a single-fraction model to a multifraction-capable knowledge-based planning model for ocular malignancies using noncoplanar treatment geometries. This approach has the potential to significantly improve planning efficiency for ocular SRS/SRT treatments. A previously validated HyperArc-based RapidPlan model for single-fraction (25 Gy) ocular SRS was adapted and further trained to handle 3- and 5-fraction prescriptions (42 Gy/3 Fx and 50 Gy/5 Fx). The knowledge-based planning (KBP) model was trained on 86 synthetic HyperArc plans and retrospectively tested on 24 datasets. Treatment planning was performed in the Varian Eclipse treatment planning system with the Acuros XB dose algorithm. Metrics included the Radiation Therapy Oncology Group conformity index, Paddick conformity index, gradient index, heterogeneity index, organ-at-risk (OAR) doses, and delivery accuracy via portal dosimetry patient-specific quality assurance (QA) and an in-house Monte Carlo second check. Across 1-, 3-, and 5-fraction ocular SRS/SRT regimens, the KBP model-based plans produced comparable target conformity, gradient, homogeneity, and coverage metrics, with no clinically meaningful differences between fractionation schemes. Differences in mean target dose (∼1%) were negligible. Most OAR constraints were met; optic nerve and lacrimal gland sparing were consistently achievable except when the OARs were included within the planning target volume, while lens and skin constraints were more challenging. Plan optimization was efficient (median ∼14 minutes), delivery times decreased with increased fractionation, and all KBP plans met patient-specific QA criteria with high pass rates during end-to-end testing and validation. These results demonstrate that a single RapidPlan model can generate high-quality ocular SRS plans across 1-, 3-, and 5-fraction schemes. The rapid generation of these KBP plans allows multiple fractionation schemes to be explored on a per-patient basis in a timely manner that would otherwise be clinically impractical with conventional manual planning.