OBJECTIVE: This study evaluates whether pre-treatment anatomical robust plan evaluation using a population-based principal component analysis (PCA) model can accurately estimate the impact of inter-fractional anatomical variations and provide meaningful advantage beyond setup-based methods in head and neck cancer proton therapy. Approach: A PCA model was built on deformation vector fields derived from planning computed tomography (CT) and weekly cone-beam CTs (CBCTs) of 20 oropharyngeal cancer patients to simulate anatomical variations. The model was applied to five additional test patients to provide anatomical scenarios. Proton therapy plans were optimized with three levels of setup robustness (4/2/0 mm). Anatomical robust plan evaluation (Ana-RE) was compared to conventional (Conv-RE) and probabilistic (Prob-RE) setup-based robust plan evaluation across different aspects: population-level trends, patient-specific accuracy and ranking of plan robustness. Reference daily dose variations were calculated using synthetic CTs derived from daily CBCTs (Daily-Ref). The analysis was repeated for a combined anatomical-setup robust plan evaluation (Ana-RE-PS) with daily synthetic CTs incorporating probabilistic setup shifts as a reference (Daily-Ref-PS). Main results: Ana-RE showed the highest agreement with Daily-Ref with a lowest median absolute errors of 0.24 percentage points for ΔV95% across clinical target volume. Across organs at risk, Ana-RE also achieved the lowest absolute errors, with 0.54 Gy for ΔDmax and 0.93 Gy for ΔDmean. Ana-RE best identified plan robustness differences, with a mean absolute rank difference of 2.3 across all plans, compared to 2.7 for Conv-RE and 2.9 for Prob-RE. When combined with setup uncertainties (Ana-RE-PS), performance remained higher than other approaches with a mean absolute rank difference of 1.20 compared to Ana-RE-PS (1.33), Conv-RE (2.12) and Prob-RE (2.67). Significance: Anatomical robust plan evaluation using a population-based model offers an estimate of the impact of anatomical variations during the treatment. Combining with setup uncertainties further extends its potential for supporting a patient individualized decision-making process prior to treatment.
Vatterodt et al. (Fri,) studied this question.