Biology-informed reinterpretation of lattice radiotherapy using non-local bystander signalling and anatomy-aware vascular sink modelling.

Objective.Lattice radiotherapy is usually evaluated using physical dose metrics such as target coverage, peak-to-valley dose ratio (PVDR), and organ-at-risk (OAR) dose, but these metrics do not explicitly represent non-local biological signalling. This study developed a hypothesis-generating computational framework to test whether synthetic head-and-neck lattice dose patterns retain the same interpretation after non-local signalling and anatomy-aware vascular uptake are included.Approach.A reaction-diffusion model propagated ROS-like and cytokine-like signals from TOPAS-derived dose distributions evaluated on a body-aware smoothed Monte Carlo dose substrate. The model was calibrated once in a one-dimensional half-field assay and transferred without geometry-specific retuning to stripe, lattice, irregular surrogate, and synthetic head-and-neck geometries. A 10-case algorithmic cohort with tumour, OAR, and vessel masks was rescored as physical-only, biology-informed without vascular uptake, and biology-informed with anatomical vascular uptake. Repeat-run uncertainty, sink falsification, smoothing-kernel sensitivity, and bounded parameter robustness analyses were performed.Main results.In the oxygen-neutral primary cohort, the cohort-mean PVDR-like ratio decreased from 1.817 under physical-dose interpretation to 1.292 without vascular uptake and 1.270 with anatomical uptake. Biology-informed reinterpretation changed ranking in 5/10 cases for no-sink versus physical-only, 6/10 for with-sink versus physical-only, and 2/10 for with-sink versus no-sink. Biology-added brainstem-adjacent burden flags occurred in 2/10 cases. In repeated runs, 65/70 sink-related endpoint shifts exceeded the combined 95% Monte Carlo plus uptake-sensitivity band, but 0/10 cases showed a noise-qualified sink-driven rank change.Significance.Non-local biological rescoring dominated reinterpretation, while vascular uptake acted as a weak secondary perturbation. The framework is hypothesis-generating and requires patient-derived RTDOSE/RTSTRUCT data, realistic vascular segmentation, and experimental validation before clinical use.