Robust Optimization Margins for Proton Therapy Using Clinically Reported Intrafraction Motion and Setup Variations on a Proteus™ One Proton System
Abstract
Purpose
Five years of field‑by‑field couch‑shift data recorded by radiation therapists on the ProteusONE™ proton system were analyzed to derive clinically relevant robust‑optimization margins and to evaluate the adequacy of existing margin guidance.
Methods
Site‑specific imaging protocol using all available modalities on the ProteusONE™ include intrafraction oblique images whenever the couch is rotated during the first three fractions and weekly thereafter. In prostate and prostate‑plus‑regional‑lymph‑node cases, imaging preceded each field in all fractions and followed each field for the first ten fractions. Shifts were grouped into four cohorts: masked immobilization, no‑mask immobilization, localized prostate, and prostate with pelvic nodes. For each cohort the mean (M), systematic standard deviation (Σ), and random standard deviation (σ) of intrafraction motion were computed. Traditional IMRT planning‑target‑volume (PTV) margins were then calculated using both the Van Herk prescription (2.5 Σ + 0.7 σ) and a simple 3σ approach on the mean couch‑shift data.
Results
Van Herk margins were 1.22‑1.26 mm (mask) and 1.25‑1.32 mm (no‑mask), with a 3‑D aggregate of 1.42 mm and 1.60 mm, respectively. For prostate‑only plans the aggregate was 2.78 mm; for prostate‑plus‑nodes it rose to 3.49 mm. The 3σ method produced consistently larger 3‑D aggregates—1.74 mm, 2.04 mm, 3.39 mm, and 4.26 mm—which are 22–28 % greater than the Van Herk estimates. Across all sites the 3σ margins offered a tighter safety buffer, notably in 3‑D space, but at the cost of higher normal‑tissue exposure. Directional Van Herk margins therefore provide the most robust plans while sparing surrounding tissues.
Conclusion
The Van Herk prescription yields smaller robust‑optimization margins than the 3σ method. Nonetheless, applying 3σ margins directionally can still preserve normal‑tissue dose while aligning robust optimization with observed uncertainties.