Imaging Differentiation of Supratentorial Ependymoma and High-Grade Glioma in Children Using VASARI Features

Supratentorial ependymomas (sEPNs) and supratentorial high-grade gliomas (sHGGs) are rare pediatric tumors with overlapping imaging features, making preoperative differentiation challenging. Accurate distinction is crucial for determining the appropriate management, guiding surgical decisions. The Visually Accessible Rembrandt Images (VASARI) feature set is a standardized MRI-based system for describing imaging characteristics of gliomas. VASARI has proved accessible, reproducible, and clinically helpful in characterizing tumor morphology. We hypothesize that a combination of imaging features can distinguish between these two tumor types. We evaluated a pediatric cohort with sEPN and sHGG to identify distinguishing imaging features, considering demographic and imaging factors. This approach aims to enhance diagnostic accuracy and improve individualized treatment planning.

This retrospective study enrolled patients younger than 21 years of age, with a histologically or molecularly confirmed sEPN or sHGG between 2000 and 2023. We evaluated 36 imaging features (54 including subcategories), incorporating the VASARI set and additional tumor characterization parameters. Univariate analysis assessed the relationships between demographic and imaging features and tumor type, followed by multivariate logistic regression. Finally, generalized binomial regression with regularization and variable selection was used to construct simplified parsimonious models of key distinguishing features for clinical use.

Forty-five patients were included, 26 sEPNs and 19 sHGGs. Sex distribution was similar between groups (61.5% female with sEPN and 78.9% with sHGG (P = .18). By univariable analysis, 16 imaging features differed significantly between tumors (P < .05), including the proportion of enhancing/nonenhancing components, calcifications, T1WI/FLAIR ratio, T2WI signal, calvarial remodeling, and involvement of specific brain regions. Multivariate analysis incorporating these features achieved 100% accuracy in differentiating the tumors (area under the curve [AUC ] = 1). A smaller parsimonious model that combined the presence of calcifications and nonenhancing margin definition accurately distinguished the tumors (AUC = 0.98). Alternatively, using enhancing and nonenhancing margin definitions also achieved high accuracy (AUC = 0.95).

Although sEPN and sHGG share overlapping imaging characteristics, a combination of 16 routine MRI features can fully differentiate them. Smaller subsets of 2 features (calcifications with definition of nonenhancing margins or the definitions of both enhancing and nonenhancing margins) also provide high diagnostic accuracy. These feature combinations improve differentiation and may support more informed treatment decisions, potentially leading to better patient outcomes.