Os02.4.a a Circuit-Based White Matter Density Neuroimaging Biomarker to Stratify, Predict, and Interpret Survival Rates in Human Glioblastoma

BACKGROUND Glioblastoma (GBM) is a lethal tumor progressing through interactions with neuronal activity, likely spreading along white matter tracts. Rather than a passive lesion, GBM appears functionally connected to distant brain regions, supported by long-range white matter connectivity. Hence, we hypothesized that tumor-implicated white matter tracts formed the structural basis for this complexity and consequently carried prognostic value. METHODS Using normative neuroimaging models, we developed and validated a lesion-tract density index (L-TDI) in two independent GBM cohorts (N=367 and N=496). The L-TDI was defined as the average fiber density within the mask containing all the tractography streamlines from the normative model that intersected the GBM. Analyses incorporated standard covariates (sex, age, surgical resection, MGMT promoter methylation) and included Kaplan-Meier curves, multivariate Cox and logistic regressions, dimensionality reduction, and a machine learning model. RESULTS First, the average tract density within the tumor-intersecting white matter map robustly stratified survival rates, reflecting the impact of widespread white matter involvement (p=0.001, p=0.007; two-sided log-rank test). Indeed, median survival times in the low L-TDI group were significantly longer in both cohorts (p<0.001; two-sided Mann-Whitney U-test). Second, the L-TDI implicitly captured tumor morphology, location, and white matter infiltration estimated from normative models. This revealed that GBM spatial distribution and its impact on survival are rooted in the anatomy of tumor-intersecting white matter pathways. Third, multivariate Cox regression demonstrated the independent prognostic value of L-TDI (HR=1.07, p=0.005; two-sided Wald’s t-test, N=617, df=5), while logistic regression showed that L-TDI significantly predicted 12-month survival (p<0.001; two-sided Wald’s z-test, N=546, df=6). Likelihood ratio tests confirmed its added contribution beyond standard clinical covariates in both the multivariate Cox (p<0.01, df=1) and logistic (p<0.001, df=1) models. These results were consistently replicated within each independent cohort. Lastly, a logistic model predicted patient death at 12 months with balanced accuracies of 0.68 and 0.65, and areas under the curve of 0.74 and 0.73 when training and testing in the two independent cohorts. CONCLUSIONS Overall, we explored the emerging paradigm of GBM as a network tumor shaped by its complex interactions with distant brain regions, highlighting how tract density imaging offers critical insights into the morphology, location, prognosis, and evolution of human GBM. FUNDING NCN (UMO-2024/53/N/NZ4/03513), EU Horizon 2020 (857533), FNP (MAB PLUS/2019/13), and Minister of Science and Higher Education (MEiN/2023/DIR/3796).