Magnetic resonance imaging features differentiate histologic and molecular subtypes of glioblastoma IDH-Wild type CNS WHO grade 4

TL;DR

This study shows that MRI features can distinguish between histological and molecular subtypes of glioblastoma, independent of surgical status.

Contribution

The study identifies MRI features that uniquely predict histological versus molecular glioblastoma subtypes using regression analysis.

Findings

Contrast enhancement, ring enhancement, and normalized ADC are MRI features that differentiate histological and molecular GBM.

The model correctly classified GBM subtypes in a validation cohort using these MRI features.

These MRI features remain predictive independent of surgical status.

Abstract

Glioblastoma IDH-wild type, CNS WHO grade 4 (GBM) can be diagnosed on the basis of histologic features (histological-GBM) or molecular features (molecular-GBM). Only few studies report neuroimaging features of GBM in its modern classification, and none have controlled for surgical status or used multiple logistic regression analysis to determine unique predictors. Our study aimed to validate MRI features that distinguish histological-GBM and molecular-GBM.

We analyzed a training cohort (n = 255) and validation cohort (n = 44) of GBM cases, classified according to the 2021 WHO Classification of Tumors of the CNS. For the training cohort, univariate and multiple logistic regression analyses determined if MRI metrics (contrast enhancement, ring-enhancement, vasogenic edema, multifocal tumor, lesion diameter, hemorrhage, number of lobes, and normalized ADC) and surgery type (biopsy vs. resection) predicted GBM-type (histological vs. molecular). A reduced multiple logistic regression model was constructed and applied to the validation dataset.

There were 231 histological-GBMs and 24 molecular-GBMs in the training cohort. Multiple logistic regression analysis including both MRI metrics and surgery type showed that contrast enhancement (OR 7.83 [95%CI: 1.23–49.68], p = 0.029), ring enhancement (OR 5.98 [95%CI: 1.09–32.93, p = 0.040), and normalized ADC (OR 0.78 [95%CI: 0.62–0.99], p = 0.039) differed between histological and molecular-GBM. Analysis of the validation dataset using the unique training dataset-derived predictor variables (contrast-enhancement, ring-enhancement, and normalized ADC) found correct classification of each histological and molecular-GBM.

Molecular and histological-GBM exhibit distinct MRI phenotypes independent of surgical status.

The online version contains supplementary material available at 10.1007/s11060-026-05431-8.

Figures