3105 - Refinement of Robust Optimization Criteria Using ?AUC Evaluation

Purpose/Objective(s): Traditional robustness metrics, such as ?D95 and worst-case dose evaluation, are influenced by nominal DVH shape, limiting usefulness for large-scale comparisons or identifying which perturbations most degrade plan robustness. ? Area Under the Curve (?AUC) provides a DVH-shape-independent metric that quantifies relative impact of individual perturbations. We evaluated ?AUC as a tool to identify low- and high-impact perturbations and to guide refinement of robust optimization criteria. We hypothesized that ?AUC-directed modification of positional and range uncertainty parameters would preserve target robustness, reduce unnecessary conservatism, and improve OAR sparing.

Materials/Methods: 10 previously approved prostate proton therapy plans (70 GyRBE to CTVp and 56 GyRBE to CTVsv in 28 fractions) were retrospectively replanned using several sets of robust optimization criteria. Prescription, beam arrangement, and objectives were held constant. For each case, ?AUC was computed for all positional (±x, ±y, ±z) and range (±) perturbations to rank their relative impact on robustness. Based on these rankings, new criteria were generated by reducing or increasing shifts based on ?AUC values. Reoptimized plans were evaluated using both classical robustness metrics (?D95) and ?AUC.

Results: Across all evaluated plans, ?AUC identified range uncertainty as the dominant contributor to CTVp degradation, while y (superior–inferior) and z (anterior–posterior) positional shifts contributed minimally. Eliminating y/z shifts entirely resulted in unacceptable CTVsv degradation under +y perturbations. However, reducing y/z shifts to ±2 mm and increasing range uncertainty to 4% maintained CTVp and CTVsv robustness comparable to the clinically used criteria while improving OAR sparing. Penile bulb D1 decreased by up to 3.5 Gy and bladder D1 by ~0.8 Gy relative to the original criteria. The modified criteria showed a similar percentage of perturbations exceeding ?D95 tolerance while reducing high-dose spill to adjacent normal tissues.

Conclusion: ?AUC effectively identified clinically relevant perturbations and provided actionable guidance for refining robust optimization criteria. A revised set of criteria using 4% range uncertainty and ±2 mm y/z positional shifts preserved target robustness and reduced OAR dose compared with our institution’s standard parameters. ?AUC shows promise as both a superior robustness metric and a tool for developing site-specific, data-driven robust optimization criteria. Evaluation of additional treatment sites is warranted to determine broader applicability.