Development of the glioma imaging complexity score (GICS): a volumetric MRI-based stratification framework

Quantitative MRI analysis and radiomics-based imaging assessment have increasingly contributed to neuro-oncological research by improving reproducibility and minimizing dependence on subjective radiological interpretation. Despite advancements in imaging analysis and visualization support technologies, standardized quantitative frameworks for classifying glioma MRI-derived complexity remain limited.

To establish a quantitative framework based on MRI data for stratifying the complexity of glioma imaging. This framework utilizes standardized imaging features derived from segmentation, obtained through the Brain Tumor Segmentation (BraTS) initiative.

A retrospective quantitative MRI analysis was performed using segmentation datasets from 1,251 glioma cases obtained from the publicly available BraTS repository. Quantitative imaging variables, including total tumor volume, enhancing tumor volume, edema volume, necrotic/non-enhancing core volume, and maximum tumor diameter, were extracted using voxel-based segmentation methods. Cases were ranked according to total tumor volume and subsequently categorized into low-, moderate-, and high-complexity groups corresponding to GICS-1, GICS-2, and GICS-3 classifications. Statistical analysis included ANOVA, Kruskal-Wallis testing, and Pearson/Spearman correlation analysis.

Tertile stratification produced three evenly distributed groups, each comprising 417 cases. The mean total tumor volume increased across GICS categories, from 35.40 ± 15.84 cm³ in GICS-1 to 162.62 ± 34.96 cm³ in GICS-3 (p < 0.001). An escalation in the GICS category was also correlated with a greater edema burden, increased maximum tumor diameter, and larger enhancing and necrotic components within the tumor. Notably, robust positive correlations were observed between total tumor volume and maximum tumor diameter (Pearson r = 0.764; p < 0.001) as well as between total tumor volume and edema volume (Pearson r = 0.861; p < 0.001). Overall, higher GICS categories demonstrated consistent quantitative differences across segmentation-derived imaging variables.

The GICS framework represents a preliminary quantitative MRI stratification model constructed using standardized segmentation datasets. The results indicate that segmentation-derived MRI variables can be organized into reproducible stratification groups, thereby laying the groundwork for prospective studies on imaging-based complexity assessment and exploratory visualization-support approaches in glioma surgery.