Preoperative prediction of Ki-67 expression and risk stratification in gliomas using multiparametric MRI and...

To investigate the feasibility of multiparametric magnetic resonance imaging (MRI)-based characterization of intratumor heterogeneity (ITH) for noninvasive prediction of Ki-67 proliferation index (PI) and assessment of its prognostic significance in gliomas.

This study included 205 patients with pathologically confirmed gliomas. Dynamic contrast-enhanced MRI and diffusion-weighted imaging were used to generate volume transfer constant (K trans ) and apparent diffusion coefficient (ADC) maps. A voxel-wise k -means clustering algorithm was applied to segment tumors into three biologically distinct intratumor habitats based on K trans and ADC values. Logistic regression and elastic net models were developed to predict Ki-67 PI. Model performance was validated through 10-fold cross-validation (CV) and two independent test cohorts, with diagnostic accuracy assessed by receiver operating characteristic curves and area under the curve (AUC). The prognostic value of habitat-derived biomarkers for progression-free survival (PFS) and overall survival (OS) was evaluated using Kaplan-Meier analysis and Cox proportional hazards models. A composite risk score was calculated for patient stratification.

Three spatial habitats were identified: H1 (hypo-vasopermeability, hypo-cellularity habitat), H2 (hypo-vasopermeability, hypercellularity habitat), and H3 (hyper-vasopermeability habitat). The elastic net model demonstrated high predictive accuracy for Ki-67 PI, with AUCs of 0.924, 0.875, 0.881, and 0.869 in the training, CV, and two test sets, respectively. Patients classified as high-risk by the risk score exhibited markedly shorter PFS (median 6.3 vs. 52.4 months) and OS (median 13.2 vs. 76.4 months) compared to low-risk patients. High-risk status was significantly associated with poorer prognosis [PFS: hazard ratio (HR) = 4.338, 95% confidence interval (CI): 2.596-7.249; OS: HR = 4.471, 95% CI: 2.572-7.772; both P < 0.001).

Multiparametric MRI-based habitat imaging effectively enables preoperative noninvasive prediction of Ki-67 expression and risk stratification in gliomas, with potential to offer insights into tumor biological behavior and inform individualized treatment planning.