Voxel-aware oxygen kinetics resolves radiation-induced DNA damage retention across LET-oxygen conditions

Objective. Hypoxic tumor subvolumes resist radiation through elevated oxygen enhancement ratios (OER), yet no computational OER model is simultaneously particle-specific, mechanistically grounded, and fast enough for voxel-scale treatment planning. We present the VOxel-Aware Oxygen Model (VOxA) to address all three requirements. Approach. An Oxygen Model (OM) encodes particle-specific LET-OER dependence through dual sigmoidal transitions constrained to increase monotonically with atomic number Z, combined with Michaelis-Menten oxygen kinetics. A Voxel-Aware (VA) extension resolves per-DSB local energy heterogeneity via a calibrated particle-specific sensitivity parameter. Calibrated on 233 OER observations from 29 sources across 10 particle types (LET = 0.2-654 keV/um); DSB coordinates from TOPAS-nBio simulations. Main results. The OM achieves $R^2 = 0.719$ and MAE = 0.300 retention OER units; theoretical OER maximum 3.32 (2.4% from measurement), bootstrap median 3.37 [3.18, 4.09]. The composite $K_{\rm fix} + K_{\rm repair} = 2.82$ mmHg is tightly constrained despite high collinearity (r = 0.935). On the Furusawa heavy-ion subset, VOxA achieves 28.4% lower survival OER MAE than the clinical standard (63.1% on helium, 24.0% on carbon) and reproduces He < C < Ne Z-ordering that universal models cannot capture. The VA extension passes 18 tests confirming sample-size-invariant within-nucleus coefficient of variation of the per-DSB retention probability. VOxA evaluates in under $10^{-3}$ ms per voxel, more than $10^6$ times faster than Monte Carlo chemistry. Significance. VOxA is the first particle-specific OER model to reproduce Z-ordering analytically at clinical planning speed, validated on the largest OER calibration dataset for this model class. Committed-break coordinates at whole-nuclear scale provide the input for inter-break topological analysis and hypoxic LET painting.