# Spatial tumour characteristics as an indirect marker of metabolic dysregulation: evaluation for non-invasive IDH-genotyping of glioma using hybrid [18 F]FET-PET/MRI

## TL;DR

This study shows that tumor characteristics seen on MRI scans can predict genetic mutations in brain tumors without invasive testing.

## Contribution

A non-invasive radiogenomic method for IDH-genotyping of gliomas using spatial tumor characteristics from MRI.

## Key findings

- IDH-mutant gliomas showed significantly lower contrast-enhancing tumor fraction (CTF) compared to wild-type gliomas.
- CTF predicted IDH genotype with high accuracy (AUC = 0.85) using standard MRI data.
- Metabolic tumor volume from [18F]FET-PET exceeded structurally apparent total tumor volume.

## Abstract

The isocitrate dehydrogenase (IDH) genotype is crucial for diagnosing and managing adult-type diffuse glioma. We investigated spatial tumour characteristics in treatment-naïve glioma using an U-Net-based CNN and evaluated associations with metabolic dysfunction and IDH genotype.

Between 2015 and 2024 patients with confirmed contrast-enhancing glioma were pre-operatively investigated using MRI or [18 F]FET PET/MRI. Automated morphometry using a U-Net-based CNN on standard MRI sequences (T1c, T1, T2, FLAIR) was performed. Contrast-enhancing tumour fraction (CTF), metabolic tumour volume (MTV), total tumour volume (TTV) were determined. Dice coefficient assessed volume intersections. Comparative and statistical analyses included non-parametric tests, ROC curves, regression, and correlation.

A total of 180 patients (male, 114; female, 66; age, M ± SD = 54 ± 15y; IDH-mutant, 63; IDH wild-type, 117) with treatment-naïve glioma were evaluated. [18 F]FET-PET metabolic activity correlated significantly with CTF (p < .05). IDH-mutant gliomas had lower CTF (p < .001) due to higher non-enhancing tumour mass (p < .001) relative to the enhancing mass, unlike IDH wild-type glioblastoma. The CTF predicted IDH genotype with high accuracy (AUC = 0.85, sensitivity 78%, specificity 90%) across datasets. Combining CTF with patient age or SUVmax further improved the classification (ΔAUC = 0.12, p = .02; ΔAUC = 0.09, p > .05). Subgroup analyses showed consistent performance across IDH-mutant subtypes. MTV from [18 F]FET-PET exceeded structurally apparent TTV (p = .033).

Spatial mapping of treatment-naïve glioma identified a non-invasive biomarker, which is linked to metabolic dysfunction and enabled robust IDH-genotype classification from standard MRI, suggesting a central role for radiogenomic assessment in adult-type diffuse gliomas prior to surgery.

The online version contains supplementary material available at 10.1007/s00259-025-07520-8.