Survival prediction in gliomas based on MRI radiomics combined with clinical factors and molecular biomarkers

To investigate the practicability of a radiomics signature combined with clinical factors and molecular biomarkers for predicting overall survival (OS) in glioma patients.

Training (n = 331) and internal validation (n = 83) sets were retrospectively collected from the Cancer Image Archive/The Cancer Genome Atlas (TCIA/TCGA), and 165 patients from our hospital for an external validation set. The least absolute shrinkage and selection operator (LASSO) was developed to select features. A radiomics model was established for predicting OS based on contrast-enhanced T1-weighted imaging (CE-T1WI) and T2 fluid attenuated inversion recovery (T2FLAIR) images. The risk stratification value of the radiomics signature was explored using Kaplan-Meier survival analysis and the log-rank test. The integrated prediction model with selected clinical factors, molecular biomarkers, and radiomics features was constructed through multivariate Cox regression analysis. Radiomics prognostic performance and benefit were assessed for all cohorts.

The radiomics signature based on the combined sequences indicated exceptional predictive ability for OS in three cohorts and stratified glioma patients significantly into high-risk and low-risk groups (P < 0.0001). A nomogram incorporating O6-methylguanine-DNA-methyltransferase (MGMT), isocitrate dehydrogenase (IDH), pathological grade, age, and radiomics signature showed excellent evaluation performance and good calibration for predicting OS in the training (C-index = 0.774), internal (C-index = 0.750), and external (C-index = 0.776) validation cohorts.

The radiomics signature demonstrates superior predictive performance for OS in glioma patients and significant subgroup risk stratification efficiency. Moreover, the comprehensive model combining clinical factors, molecular biomarkers, and radiomics features further achieves a robust assessment of survival prognosis.