Dose-guided positioning optimized proton target coverage within three minutes

KEY POINTS

• The study included 45 head and neck cancer patients treated with definitive IMPT to a primary CTV dose of 70 Gy(RBE) and an elective CTV dose of 54.25 Gy(RBE) in 35 fractions. Each patient had five to seven weekly repeat CT scans; nine underwent offline plan adaptation.
• The dose-guided positioning algorithm combined the GPU-accelerated Moqui Monte Carlo engine with gradient-descent optimization. It searched for the translational shift maximizing V95% of both targets, with greater weight assigned to the primary CTV.
• Moqui was commissioned against the institution’s clinical planning system across 45 plans. Three-dimensional gamma pass rates using 2%/2 mm criteria were at least 99.1%, supporting its use for repeated optimization calculations.
• Six initial positions were tested for every repeat CT, each generated within ±7 mm of the conventional anatomy-based alignment. The optimization allowed up to eight iterations and applied a 1 mm expansion around the CTVs during target-coverage evaluation.
• Compared with anatomy-based positioning, mean V95% differences were +0.1% ± 0.3% for the 70 Gy(RBE) CTV and −0.1% ± 0.4% for the elective CTV, showing that target coverage was generally preserved.
• In scans where anatomy-based positioning failed to achieve the institutional target of V95% ≥98%, dose-guided positioning improved primary CTV coverage by 1.0% ± 0.8%. The average optimization time was 149 ± 51 seconds.
• The objective function considered target coverage but not organ-at-risk dose. Testing used diagnostic-quality repeat CTs and known contours rather than daily cone-beam CT or synthetic CT, and the algorithm optimized translations without evaluating rotational corrections, contour uncertainty, or prospective treatment delivery.

CLINICAL TAKEAWAY

SOURCE

Medical Physics