NIRA-10 RADIATION NECROSIS VERSUS TUMOR IN BRAIN METASTASES: STRUCTURAL AND PHYSIOLOGIC HABITAT IMAGING WITH HISTOPATHOLOGICAL GROUND TRUTH
Jieun Park, Guowen Shao, Shivani Baisiwala, Nakyoung Kim, Amelia Tan, Andrea Liang, Zexi Wang, Francesco Sanvito, Gianluca Nocera, Catalina Raymond, Vien Le, Hosung Kim, Noriko Salamon, Whiteny Pope, Won Kim, Benjamin Ellingson, Jingwen YaoAbstract
Background
Differentiating tumor recurrence from radiation necrosis (RN) after radiation therapy remains a critical clinical challenge in brain metastasis (BM) management. Tumor habitat analysis captures intratumoral heterogeneity via unsupervised voxel clustering of multiparametric MRI, but external validation with histopathological ground truth has been limited. We aimed to validate established MRI-based tumor habitat analysis for distinguishing tumor from RN in an independent cohort with histopathological ground truth.
Methods
This retrospective study included 224 patients (188 BM, 36 RN) with pathologically confirmed diagnoses who underwent structural MRI (contrast-enhanced T1-weighted, T2-weighted) and physiologic MRI (apparent diffusion coefficient, normalized cerebral blood volume). An established unsupervised clustering model was applied to generate structural habitats (enhancing, solid low-enhancing, nonviable) and physiologic habitats (hypervascular, hypovascular cellular, nonviable). Habitat volumes and volume fractions were compared between groups. Logistic regression and receiver operating characteristic analysis evaluated the ability to differentiate tumor and RN. Composite habitat scores integrating structural and physiologic habitats were also developed.
Results
Metastatic tumors demonstrated significantly larger volumes and fractions of solid low-enhancing (22.3% vs. 11.2%, P=.012) and hypervascular habitats (18.1% vs. 3.1%, P=.003), while RN exhibited higher nonviable tissue fractions on both structural (67.9% vs. 49.5%, P<.001) and physiologic MRI (31.0% vs. 20.4%, P=.003). On structural MRI, solid low-enhancing fraction >24.4% and nonviable fraction ≤62.2% predicted tumor. On physiologic MRI, hypervascular fraction >3.3% and nonviable fraction ≤9.7% predicted tumor. Physiologic MRI habitat score achieved AUC=0.81, outperforming structural MRI (AUC=0.71). The combined habitat score yielded the highest performance (AUC=0.84; sensitivity 68.1%, specificity 83.3%), with consistent results in patients with prior radiation therapy.
Conclusions
MRI-based tumor habitat analysis provides an externally validated, reproducible approach to distinguish tumor recurrence from RN in BM after radiation therapy. Combined structural and physiologic habitat scoring offers the best diagnostic performance and may guide post-radiation management decisions.