Preoperative MRI radiomics predicts IL13Rα2 expression and survival in glioblastoma
TL;DR
The derived Rad-score demonstrated prognostic relevance and remained significantly associated with overall survival in Cox regression analyses, supporting its potential as a noninvasive imaging biomarker for preoperative molecular characterization and risk stratification.
Abstract
To develop and validate a preoperative contrast-enhanced T1-weighted imaging (CE-T1WI)-based radiomics model for noninvasive prediction of interleukin-13 receptor alpha 2 (IL13Rα2) expression and prognostic stratification in glioblastoma (GBM). A total of 422 patients with 2016 WHO grade IV GBM from The Cancer Genome Atlas, the Chinese Glioma Genome Atlas, and an institutional cohort were included. Public datasets were used to assess IL13Rα2 expression, prognostic significance, and tumor microenvironment associations. For radiomics analysis, 92 patients with paired transcriptomic data and CE-T1WI were used for model development and internal validation. A total of 1197 radiomics features were initially extracted from CE-T1WI images, of which 1145 features with intraclass correlation coefficients (ICCs) > 0.75 were retained after interobserver reproducibility assessment. Features were further selected using least absolute shrinkage and selection operator regression and integrated into a radiomics score (Rad-score). Logistic regression, random forest, and ExtraTrees classifiers were constructed. An independent institutional cohort of 36 patients with matched CE-T1WI and formalin-fixed paraffin-embedded specimens was used for exploratory cross-platform external validation, with IL13Rα2 immunohistochemistry serving as the protein-level reference standard. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), and clinical utility was evaluated by decision curve analysis. The prognostic value of the Rad-score was assessed using univariable and multivariable Cox regression analyses. IL13Rα2 was overexpressed in GBM and independently associated with worse overall survival. In the institutional cross-platform validation cohort, the highest-AUC radiomics model achieved an AUC of 0.801 for predicting IL13Rα2 expression. A high Rad-score was associated with shorter overall survival (HR = 2.03, 95% CI: 1.24–3.34, P = 0.005). In multivariable Cox regression analysis, the Rad-score remained significantly associated with overall survival after adjustment for clinical variables. In addition, the Rad-score was positively associated with IL13Rα2 expression and with exploratory immune-related features. CE-T1WI-based radiomics shows potential for the noninvasive prediction of IL13Rα2 expression in GBM. The derived Rad-score demonstrated prognostic relevance and remained significantly associated with overall survival in Cox regression analyses, supporting its potential as a noninvasive imaging biomarker for preoperative molecular characterization and risk stratification.