BLUE RIBBON POSTER IMAGING: A Novel Graph-Based Characterization of Peritumoral Vascular Architecture for Pre-Treatment Assessment of Radiation Necrosis Risk In Brain Metastases
Abstract
Purpose
Pre-treatment heterogeneity in vascular architecture may contribute to differential susceptibility to radiation necrosis (RN), a dose-limiting toxicity of stereotactic radiosurgery (SRS). Using routine pre-treatment MRI, we introduce a novel graph-based analysis to quantitatively characterize peritumoral vascular networks and evaluate associated RN risk.
Methods
76 brain metastases from 16 patients treated with SRS were analyzed. RN was confirmed in 24 lesions by biopsy or 6-month follow-up imaging. Pre-treatment T1-weighted and contrast-enhanced T1 MRI were processed to extract vascular structures using subtraction-based enhancement and centerline skeletonization. Graph-based representations of vascular topology (VasGraph) were then constructed by tracing vascular skeletons to define nodes and edges within each tumor neighborhood (GTV with 1cm peritumoral expansion). Seven sets of graph-derived metrics characterizing vascular network capacity and efficiency were computed for each VasGraph and for GTV-restricted subgraphs. Univariate group comparisons and logistic regression were used for individual feature assessment, followed by multivariate Lasso regression for feature selection and improved RN risk prediction.
Results
As expected, RN-prone tumors exhibited significantly reduced betweenness in the tumor neighborhood (BEC, AUC = 0.76, p = 0.001), along with reduced global efficiency (GE, AUC = 0.85, p < 0.001) and increased characteristic path length (CPL, AUC = 0.81, p = 0.001) within GTV subgraphs, indicating reduced network capacity and disrupted vascular topology. Multivariate Lasso regression incorporating VasGraph metrics further improved RN prediction, achieving an AUC of 0.89, with an accuracy of 0.90, sensitivity of 0.93, and specificity of 0.90 under five-fold cross-validation.
Conclusion
Using routine clinical MRI, we present a novel VasGraph pipeline for quantitative, interpretable characterization of peritumoral vascular topology to assess pre-treatment RN susceptibility. Graph analysis captures vascular architectural disruption associated with RN development, providing insight into vascular contributions to RN risk. These metrics enable lesion-specific vascular profiling to support pre-treatment stratification and personalized treatment planning.