A Mechanistic Model of Brain Necrosis Progression Based on Vascular Heterogeneity

Brain radionecrosis (RN) is a significant late toxicity of radiation therapy, yet its progression remains challenging to predict because of patient-specific factors. This study develops a mechanistic model to simulate RN expansion focusing on vascular heterogeneity.

A 3-dimensional cellular automaton (CA) model was developed to simulate RN progression, based on the assumption that vascular heterogeneity drives its spatial dynamics. Patient-specific vasculature maps were generated by registering a synthetic brain phantom to magnetic resonance imaging-derived segmentations. Microvessel length density (L_d) was estimated to account for regional vascular heterogeneity. The model parameters-RN progression rate (k) and necrotic neighborhood threshold (ρ_t)-were inferred using sequential Monte Carlo approximate Bayesian computation. Probability risk maps were validated against follow-up (FU) imaging from 3 independent cases, with voxelwise agreement assessed using receiver operating characteristic analysis.

The model successfully predicted RN expansion patterns, achieving area under the curve values of 0.87 to 0.95 in validation cases. Simulated necrotic regions exhibited anisotropic expansion influenced by local vascular density, supporting the vascular hypothesis. Patient-specific posterior distributions for progression rate reflected wide interpatient variability, whereas the necrotic neighboring effect had a narrower range. The model consistently identified high-risk voxels, with predicted necrotic regions overlapping observed RN in FU imaging.

This study presents a mechanistic model that integrates vascular heterogeneity to predict RN progression, providing interpretable, patient-specific risk maps. It extends RN evolution modeling beyond dose-based metrics, potentially aiding in refining treatment planning and adaptive FU strategies to minimize radiation-induced toxicity.