Risk stratification prediction of glioblastoma based on multi-sequence MRI radiomics analysis
Objective:To develop a glioblastoma(GBM)overall survival(OS)prediction model using multi-sequence MRI radiomics method.Materials and Methods:This study retrospectively collected data from 309 patients with GBM in the TCIA/TCGA(The Cancer Imaging Archive/The Cancer Genome Atlas)public database,and extracted 10 128 radiomics features from preoperative post-contrast enhanced T1-weighted(T1CE)and T2-weighted fluid attenuation inversion recovery(T2 FLAIR)sequences for three regions of interest:necrosis area,tumor area,and edema area.Correlation analysis,principal component analysis(PCA)for dimensionality reduction,and least absolute shrinkage and selection operator-cox proportional-hazards(LASSO-Cox)regression were used to screen radiomics features significantly related to OS and calculate a Risk-score as a radiomics signature.Kaplan-Meier(KM)survival analysis and Log-rank test were used to compare the survival differences between high-risk and low-risk groups.A clinical-radiomics combined model and nomogram were constructed using multivariate Cox proportional-hazards(Cox)regression and evaluated using concordance index(C-index),which was compared with the clinical model.Results:Based on the 16 radiomics features selected from the training set,a Risk-score was calculated and patients were divided into high-and low-risk groups based on this Risk-score,using both the training and testing sets.Log-rank testing showed a significant difference in survival probability between the high-and low-risk groups.Univariate Cox regression identified Risk-score,age,and O6-methylguanine-DNA methyltransferase(MGMT)status as significant risk factors for OS in GBM.Multivariate Cox regression was used to build clinical model and a clinical-radiomics combined model,and it was found that the clinical-radiomics combined model(training set:C-index=0.768,testing set:C-index=0.724)outperformed the radiomics model(training set:C-index=0.744,testing set:C-index=0.710)and the clinical model(training set:C-index= 0.659,testing set:C-index=0.653).Conclusions:Radiomics signature can serve as independent prognostic factor for OS of GBM,and the combined model constructed by combining clinical pathological information and radiomics signature can better assist in risk stratification and survival prediction of GBM,which holds significant clinical value.
glioblastomamagnetic resonance imagingoverall survivalradiomicsprediction model
国家科技期刊平台
NSTL
万方数据
