Implementing Plan Robustness Evaluation In Multi-Criteria Optimization for Prostate HDR Brachytherapy
Abstract
Purpose
HDR prostate brachytherapy (HDR-PBT) treatment plans are subject to multiple uncertainties that can significantly impact delivered dose distributions. This work implements a comprehensive robust evaluation framework in the context of multi-criteria optimization (MCO) that explicitly accounts for major uncertainty sources in HDR-PBT and provides quantitative metrics to assess robustness as function of the MCO plan pool size.
Methods
A robust evaluation framework was applied to HDR-PBT plans generated under MCO, incorporating fourteen clinically relevant uncertainty sources, including needle positioning, source positioning, anatomical deformations, and dosimetric variations. Uncertainty propagation was performed using Monte Carlo–based probabilistic sampling of all parameters to generate 2000 scenario-based dose distributions per MCO plan, and related probabilistic dose–volume histograms (DVH). Plan robustness was quantified using pass-rate metrics, defined as the percentage of uncertainty scenarios satisfying predefined clinical dose–volume constraints for target coverage (D90) and organ-at-risk (OAR) sparing for the urethra and rectum. For comparison, a geometric robustness metric suggested by Poder et al. was also computed.
Results
Maximum robustness metrics exhibited a steep increase for small MCO plan pool sizes before reaching a plateau. The standard RTOG-based metric achieved the highest robustness value of 90.0%, while institutional metric (INST) plateaued at 83.0%. The more stringent variants (RTOGp and INSTp) demonstrated lower maximum robustness, stabilizing at 67.0% and 47.5%, respectively. The normalized geometric metric showed similar behavior. Of note, all metrics achieved approximately 95% of their maximum value within the first 100 Pareto-optimal plans, suggesting diminishing returns for larger pool sizes.
Conclusion
This study adds a comprehensive robustness evaluation framework, explicitly incorporating major sources of delivery uncertainty, to MCO. It provides quantitative insight into the selection of robust plans contained within an MCO plan pool, beyond conventional nominal metrics. These tools could support uncertainty-aware plan quality evaluation and decision-making in HDR-PBT.