Work

Voxel-wise Radiomics Habitat Modeling of Multiparametric Magnetic Resonance Imaging for Isocitrate Dehydrogenase Status Prediction and Survival Stratification in Diffuse Glioma

Year: 2026
Type: preprint
Study design: Observational Study
Abstract: Purpose To develop and externally validate a voxel-wise probabilistic multiparametric MRI habitat framework for preoperative isocitrate dehydrogenase classification in diffuse glioma and to explore its association with overall survival. Material and Methods This retrospective multicenter study included 269 adults with pathologically confirmed diffuse glioma, comprising 174 training and 95 external-validation patients. Preoperative MRI included contrast-enhanced T1-weighted, T2-weighted, T2-weighted fluid-attenuated inversion recovery, apparent diffusion coefficient, and cerebral blood flow images. Tumor and peritumoral edema were segmented using nnU-Net v2. Sliding-window radiomics generated voxel-wise feature maps, and Gaussian mixture models constructed probabilistic habitats. Habitat metrics, tumor–edema differential features, and clinical variables were used for IDH classification. Exploratory survival analysis used a separate imaging-only signature derived from 13 retained imaging predictors and evaluated by Kaplan–Meier and Cox regression analyses. Results The imaging-only habitat model (TEΔ-ALL) achieved an external-validation an area under the receiver operating characteristic curve of 0.921. The combined clinical–imaging model (Clin+TEΔ-ALL) achieved the highest discrimination (AUC, 0.955; 95% CI, 0.905–0.989), outperforming the clinical-only model (AUC, 0.660), with favorable calibration and greater net benefit. SHAP identified perfusion- and diffusion-related entropy metrics as major contributors. In the external survival cohort, the imaging-only signature stratified overall survival (log-rank P < 0.0001) and remained independently associated with survival after adjustment (hazard ratio, 4.225 per standard-deviation increase; 95% CI, 2.355–7.580; P < 0.001). Conclusion Voxel-wise probabilistic MRI habitat modeling captured spatial tumor heterogeneity and showed potential for noninvasive IDH assessment and survival risk stratification in glioma.
Language: English
Field: Medicine
Open Access status: green
OpenAlex ID: W7204133167
Cites: 37
Cited by:
Related to: 10