Rethinking Robustness In Head-and-Neck IMPT: Systematic Versus Random Setup Uncertainty
Abstract
Purpose
To compare head-and-neck (HN) IMPT robustness using a conventional systematic setup uncertainty model versus a probabilistic random (interfractional) setup uncertainty model.
Methods
Two IMPT plans (70Gy(RBE) in 35 fractions) for a bilateral HN cancer patient were generated in RayStation using identical beam parameters, planning objectives, and objective weightings; no plan scaling was performed to force CTV coverage. IMPT-R used robust optimization with probabilistic random per-fraction setup uncertainty and ±3.5% range uncertainty. Random setup error was modeled as a 3D Gaussian distribution (σ=1 mm), discretized on a 5×5×5 grid and truncated at 3σ; robust evaluation used a 200-scenario per-fraction with ±3.5% density uncertainty. IMPT-S used robust optimization and evaluation with systematic setup uncertainty (±3 mm) and ±3.5% density uncertainty. Nominal target coverage was summarized using D95 and D98, while robustness for targets focused on voxelwise-worst D95. OAR endpoints included mean dose (Dmean) to the oral cavity, contralateral-parotid, and ipsilateral-parotid, reported as nominal and voxelwise-worst values.
Results
Across all CTV dose levels (CTV_7000, CTV_5950, and CTV_5600), nominal D95/D98 values were comparable between IMPT-R and IMPT-S. Under robustness evaluation, voxelwise-worst CTV D95 reductions from nominal ranged from 0.37–0.52 Gy(RBE) for IMPT-R, compared with larger reductions of 1.37–1.70 Gy(RBE) for IMPT-S. Relative increases in OAR Dmean from nominal to voxelwise-worst were also larger for systematic than random setup uncertainty, at approximately 21% vs 9% for the oral cavity, 22% vs 5% for the contralateral parotid, and 19% vs 4% for the ipsilateral parotid.
Conclusion
This novel comparison shows that systematic setup uncertainty causes larger degradations in voxelwise-worst CTV D95 and higher robust OAR doses than random setup uncertainty in HN IMPT. While systematic robustness remains the clinical standard, these results suggest that random setup error robustness merits further study as a complementary approach with potential to reduce OAR dose without compromising target coverage.