3046 - Rapid Parallel Auto-Planning to Objectively Determine Optimal Cone vs. MLC Strategy in CyberKnife S7 Radiosurgery
Presenter(s)
Y. Lai, X. Hu, L. R. Kleinberg, K. J. Redmond, X. Jia, and A. Qin; Department of Radiation Oncology and Molecular Radiation Sciences, Johns Hopkins University, Baltimore, MD
Purpose/Objective(s): CyberKnife radiosurgery supports both multileaf collimator (MLC)–based and fixed cone–based delivery, each exhibiting distinct dosimetric tradeoffs. Cone-based plans typically provide sharper penumbra and steeper dose gradients for small targets, whereas MLC-based plans offer greater modulation flexibility and potential efficiency advantages. However, generating high-quality plans for both modalities is time- and labor-intensive, making routine dual-modality comparison impractical in busy clinical settings. Consequently, collimator selection is often based on empirical heuristics—such as target size—rather than patient-specific dosimetric evidence, introducing variability and potential suboptimal decision-making. This study develops a rapid parallel automated planning framework to simultaneously generate cone- and MLC-based plans, enabling objective, evidence-based collimator selection while substantially reducing planning burden.
Materials/Methods: A RayStation scripting–based automated workflow was implemented to encode CyberKnife planning heuristics, including collimator size selection, node arrangement strategies, dose-shaping structures, and iterative optimization sequencing. Clinical goals were automatically translated into optimization objectives, and adaptive weight escalation was applied to progressively satisfy target coverage and organ-at-risk (OAR) constraints. For each case, multiple candidate plans with varying segment-per-node limits and node configurations were generated automatically in parallel for both cone and MLC delivery. The optimal plan for each modality was selected based on predefined dosimetric and delivery efficiency criteria. Planning time, conformity index (CI), gradient index (GI), homogeneity index (HI), OAR maximum doses were evaluated.
Results: The workflow was evaluated on 5 spine cases (PTV 46.3–439.9 cm³) and 5 brain metastases cases (PTV 1.3–15.6 cm³). Average planner time was reduced to <25 minutes compared to 0.5–3 days for manual planning. Automated cone plans achieved mean CI/GI/HI of 1.9/4.9/1.8, while automated MLC plans achieved 1.9/7.2/1.5. Corresponding clinical plans demonstrated 2.1/6.0/1.6. Maximum doses to spinal cord, esophagus, brainstem, and cochlea were comparable across modalities. Notably, automated plans reduced average optic maximum dose from 702 cGy (clinical) to approximately 440 cGy.
Conclusion: This rapid parallel auto-planning framework enables objective, reproducible comparison between cone- and MLC-based CyberKnife strategies while reducing manual planning time by more than an order of magnitude. The approach supports evidence-based collimator selection tailored to patient-specific geometry and reduces inter-planner variability, providing a practical pathway toward standardized and efficient CyberKnife radiosurgery planning.