3713 - Benchmarks for Proton Robust Evaluation Using Population Based Analytics

Purpose/Objective(s): Proton pencil beam scanning (PBS) plans are sensitive to uncertainties in patient setup, anatomy, and density. Robust optimization addresses these uncertainties during inverse planning, but robustness evaluation remains subjective and relies on interpretation of select dosimetric endpoints (e.g., D1, D95, Dmin, Dmax). This study introduces a population-based robust analysis using a large-scale database of historical PBS plans with systematic perturbations. This analysis establishes quantitative thresholds for plan acceptability and the first large-scale benchmark for proton plan robustness.

Materials/Methods: A Python script was developed to automate the calculation of PBS perturbed dose distributions. The script recalculates dose distributions under user-defined perturbations, including shifts, rotations, angles, and density uncertainty. Perturbed DVH data for specified structures were stored in a centralized repository over six years across 7,370 treatment plans generating a database of over 70,000 DVHs. To quantify robustness, a population-based tolerance framework was established. DVHs from all robust scenarios for matching patient groups were combined into a composite population-based histogram for robustness assessment. The acceptable band (1s) was defined as the range within which the perturbed dose remained within expected variations for the population, while the action level (2s) indicated deviations significant enough to warrant additional evaluation. Both were calculated relative to the nominal unperturbed dose for the original plan in the robust evaluation.

Results: A total of 26 hypofractionated prostate patients with 348 perturbations for the CTV_70CGE, 31 hypofractionated breast patients with 372 perturbations to the CTV_42.5CGE, and 48 head and neck (H&N) patients with 935 perturbations to the CTV_50CGE were analyzed. For the prostate CTV, the acceptable band was 1.0% of the prescription, and the action level was 2.0%. Among 348 perturbations, 23% exceeded the acceptable band, and 6.6% exceeded the action level. The most frequent cause of threshold exceedance was density uncertainty, followed by roll rotations. For the breast CTV, the acceptable band was 1.7% of the prescription, and the action level was 3.4%. Of the 372 perturbations, 17.2% exceeded the acceptable band, occurring most frequently in the anterior-posterior direction, followed by density uncertainties. For the H&N CTV, the acceptable band was 0.9% of the prescription, and the action level was 1.9%. Of the 935 perturbations, 20.5% exceeded the acceptable band, occurring most frequently in the anterior-posterior direction, followed by the right-left direction.

Conclusion: This study developed population-based analytics for evaluating the robustness of proton plans. By analyzing a large dataset of perturbed DVHs, clinically relevant tolerance and action levels based on statistical thresholds can be established.