Retrospective analysis evaluates risk stratification of adult diffuse gliomas using DSC-MRI, indicating hemodynamic features as predictive biomarkers.
Key Points
The study aims to evaluate habitat-based radiomics models from DSC-MRI to identify aggressive phenotypes of diffuse gliomas tied to histologic-molecular risk.
Retrospective analysis of 197 adult patients with histopathologically confirmed diffuse gliomas.
Multiparametric MRI data preprocessed for tumor and peritumoral edema segmentation, with habitat identification using K-means clustering.
Random forest classifiers developed to predict high-risk molecular subtypes validated in an internal testing cohort.
Habitat-based models showed superior predictive ability over whole-region analyses (AUC 0.949 vs. 0.931, p = 0.013).
Hemodynamic features predicted aggressive subtypes better than anatomical sequences in tumor habitats (AUC 0.944 vs. 0.895) and edema habitats (AUC 0.932 vs. 0.819).
Optimal model made from hemodynamic features alone achieved an AUC of 0.949.