Inside the Brain Tumor: What MRI-Defined Habitats Exist Within Glioblastoma, How Are They Spatially Organized, and What Molecular Processes Are Associated with Them?
Glioblastoma exhibits substantial spatial heterogeneity, but MRI studies often emphasize tumor extent rather than the organization of imaging phenotypes. We investigated whether multiparametric MRI identifies reproducible habitats and whether habitat abundance and spatial architecture provide distinct information. Preoperative T1, contrast-enhanced T1, T2, and FLAIR scans with manual segmentations from 97 TCGA-GBM patients were analyzed. Unsupervised clustering of normalized voxel intensities identified two habitats. An Interface Ratio quantified the observed H1-H2 boundary relative to abundance-preserving spatial randomization. Molecular associations were assessed using matched Affymetrix U133A expression data and seven prespecified transcriptional programs. H2 showed higher T2 and FLAIR signal, while habitat distributions corresponded poorly to conventional segmentation compartments. Among 66 mixed-habitat tumors, Interface Ratio varied substantially and was independent of H2 abundance (rho = 0.074, p = 0.555). Molecular data were available for 89 patients, including 60 mixed-habitat tumors. Higher H2 abundance was nominally associated with hypoxia-program scores after adjustment for Interface Ratio (beta = 0.337, 95% CI 0.089-0.584; p = 0.0077), but not after multiple-testing correction (FDR = 0.108). No Interface Ratio association survived correction. Habitat composition and spatial organization represent two complementary dimensions of glioblastoma heterogeneity. The H2-hypoxia relationship remains exploratory and requires independent, spatially resolved validation.