Automated Quantitative Analysis of Enhancing and Peritumoral Cerebral Blood Volume for Differentiating Glioblastoma From Central Nervous System Lymphoma.

[OBJECTIVE] To quantitatively compare cerebral blood volume (CBV) in contrast-enhancing tumor areas and nonenhancing peritumoral fluid-attenuated inversion recovery (FLAIR) hyperintense regions between glioblastoma and central nervous system lymphoma (CNSL), and to assess the incremental diagnostic value of peritumoral CBV beyond enhancing tumor CBV.

[METHODS] The study included 34 patients with histopathologically confirmed glioblastoma (n=22) or CNSL (n=12) who underwent pretreatment magnetic resonance imaging, including FLAIR, dynamic susceptibility contrast perfusion imaging, and contrast-enhanced T1-weighted imaging. Automated regions of interest (ROIs) were defined for enhancement areas (EAs) and nonenhancing peritumoral FLAIR abnormalities (PFAs). Quantitative indices included the median CBV within EAs (CBVEA) and the 95th percentile CBV within PFAs (CBVPFA). Intergroup differences were assessed, and diagnostic performance was evaluated using univariate and multivariate logistic regression models incorporating CBVEA and CBVPFA, receiver operating characteristic (ROC) analysis, and likelihood ratio testing (LRT).

[RESULTS] Both CBVEA and CBVPFA were significantly higher in the glioblastoma group than in the CNSL group [CBVEA: 4.90 (4.14-5.58) vs. 2.84 (2.04-3.21) mL/100 g, P<0.0001, CBVPFA: 5.67 (3.75-8.08) vs. 4.10 (3.00-5.23) mL/100 g; P=0.029]. CBVEA showed the highest discriminatory performance in univariate analysis, whereas CBVPFA demonstrated a more modest association with tumor type. Although the AUC of the combined model was not significantly different from that of the CBVEA model alone [CBVEA: 0.913 (0.816-1.000) vs. combined model: 0.932 (0.848-1.000), P=0.602], the combined model showed a significant improvement in model fit according to LRT (χ2=4.1, P=0.043).

[CONCLUSIONS] Automated ROI-based analysis demonstrated significant differences in peritumoral CBV between glioblastoma and CNSL, and adding peritumoral CBV to the logistic regression model significantly improved overall model fit for differentiating between these entities beyond enhancing tumor CBV alone.