MRI radiomics moderately predicted VEGFA expression in lower-grade glioma

KEY POINTS

• The study used TCGA data from 412 patients with WHO grade II-III glioma for molecular and survival analyses. A matched TCIA radiomics cohort contained 149 contrast-enhanced T1-weighted MRI cases, divided into 90 training and 59 internal-validation patients.
• High VEGFA expression was associated with inferior overall survival. On multivariable Cox analysis, VEGFA remained independently associated with mortality with an HR of 2.545 (95% CI 1.422-4.555, p = 0.002); tumor grade did not retain significance in that model (p = 0.059).
• The adverse association persisted across several exploratory subgroups. High VEGFA was associated with worse survival both in patients aged <40 years (HR 2.186, p = 0.035) and those aged >40 years (HR 2.239, p = 0.024), with no significant interaction across the evaluated clinical variables.
• Of 107 extracted radiomic features, 95 achieved interobserver ICC ≥0.80. RFE and mRMR ultimately selected seven features for the logistic-regression model and five for the more parsimonious AIC-optimized model.
• Predictive performance was moderate. The standard logistic model achieved AUCs of 0.728 in training and 0.704 in internal validation, while the AIC-optimized model achieved 0.725 and 0.718, respectively. The two approaches did not differ significantly in either training (p = 0.929) or validation (p = 0.719).
• Higher VEGFA expression correlated with M1 and M2 macrophages, T follicular-helper cells, and neutrophils, while monocyte infiltration was inversely correlated. High-VEGFA tumors also showed enrichment of Wnt and B-cell receptor signaling pathways.
• The model was derived from retrospective public datasets, lacked independent external validation, and retained potential scanner and acquisition heterogeneity despite standardized preprocessing. The authors explicitly state that it should not replace molecular testing or guide treatment decisions.

CLINICAL TAKEAWAY

SOURCE

Medical Physics