Dose-Gated LET Modulation of Mandibular Osteoradionecrosis In Proton Therapy Revealed By Iso-Risk Interaction Space
Abstract
Purpose
Dose–volume metrics alone incompletely explains mandibular osteoradionecrosis (ORN) following proton therapy. LET-based aggregate approaches such as dose–LET volume histograms (DLVH) implicitly assume uniform LET contribution across dose levels. This study investigates whether ORN exhibits a dose-gated, LET-modulated risk structure that can be resolved using an iso-risk interaction framework beyond one-dimensional aggregation.
Methods
Voxel-level mandibular dose and dose-averaged LET (LETd) were analyzed from six consecutive head-and-neck proton therapy patients with clinically adjudicated ORN. Dose was normalized per patient (D/Dmax), and analysis was restricted to high-dose mandibular regions using dose gates (D/Dmax ≥ 0.5–0.7), corresponding approximately to ~35-50 Gy and above. ORN risk was modeled using logistic regression incorporating dose, LETd, and a dose×LET interaction term. Iso-risk contours were derived to visualize joint dose–LET geometry. To reduce dose imbalance, nearest-neighbor dose matching was performed within high-dose regions. Model discrimination was assessed using pooled voxel-level ROC analysis across all patients.
Results
In pooled ROC analysis, dose_norm alone achieved AUC = 0.773 and LET alone achieved AUC = 0.534, whereas the combined dose_norm + LET model improved discrimination to AUC = 0.845, demonstrating added predictive value from LET when conditioned on dose. Within dose-gated mandibular regions (D/Dmax ≥ 0.6), ORN-associated voxels formed a triangular joint risk structure in dose–LET space. Iso-risk contours demonstrated that the LET required to reach a fixed modeled risk level decreased with increasing dose, indicating dose-dependent LET modulation rather than uniform LET contribution.
Conclusion
Mandibular ORN in proton therapy exhibits a dose-gated, LET-modulated joint risk structure that cannot be captured by uniform LET weighting or one-dimensional aggregation such as DLVH. Iso-risk interaction space reveals that progressively lower LET is sufficient to reach equivalent modeled risk as dose increases into the high-dose regime. These findings support LET-aware plan evaluation specifically within high-dose mandibular subvolumes, refining risk assessment beyond conventional DVHs.