Intratumoral and Peritumoral Habitat Radiomics on Multiparametric MRI for Preoperative Prediction of 1-Year...

This study aimed to develop a preoperative multiparametric magnetic resonance imaging (MRI)-based habitat radiomics model integrating intratumoral and peritumoral heterogeneity to predict 1-year progression-free survival (PFS) status in glioblastoma (GBM).

This retrospective multicenter study included 238 patients with pathologically confirmed GBM from three centers, divided into a training cohort (n = 127), an internal test cohort (n = 54), and an external validation cohort (n = 57). Preoperative multiparametric MRI (contrast-enhanced T1-weighted imaging, T2-weighted imaging, and diffusion-weighted imaging) was used for the segmentation of intratumoral, 10-mm and 20-mm peritumoral, and peritumoral edema regions. Voxel-wise K-means clustering was applied within each region to identify imaging subregions ("habitats") with similar characteristics. Following feature selection, intratumoral radiomics and habitat models were constructed using five algorithms. The best-performing algorithm was used to develop peritumoral habitat models, and a final combined model was generated by integrating the intratumoral and peritumoral habitat models with the clinical model. Model performance was evaluated using receiver operating characteristic analysis [area under the curve (AUC)], calibration curves, and decision curve analysis, with DeLong tests for AUC comparisons, integrated discrimination improvement/net reclassification improvement (IDI/NRI) for incremental value, and SHapley Additive exPlanations for interpretability.

The intratumoral habitat model (using support vector machine) outperformed conventional intratumoral radiomics (external AUC, 0.806 vs 0.702). Among peritumoral models, the 10-mm habitat model performed best (external AUC, 0.734). The combined model achieved the highest discrimination (AUCs: 0.934 training, 0.910 internal test, and 0.874 external validation), with good calibration and the greatest net benefit on decision curve analysis; DeLong tests and IDI/NRI confirmed significant incremental value over the clinical and conventional radiomics models.

Multiparametric MRI-based habitat analysis provides a more granular characterization of intratumoral and peritumoral heterogeneity in GBM. The integrated habitat radiomics combined model showed strong prognostic utility for 1-year PFS status, supporting identification of high-risk patients, optimization of follow-up strategies, and individualized treatment intensification.