Construction And Evaluation Of Multiparametric MRI-Based Radiomics Survival Stratification Model In Patients With Glioblastoma

ObjectiveBased on the radiomics features extracted from preoperative multiparametric Magnetic Resonance Imaging(MRI)in patients with glioblastoma(GBM),construct and evaluate a preoperative diagnostic tool,the radiomics nomogram model,to predict survival stratification in GBM patients receiving standardized therapy for clinical decision-making.MethodsThe clinical data of 125 patients with GBM who underwent surgery at Neurosurgery of the First Affiliated Hospital of Zhengzhou University from January 2018 to January 2020 and were diagnosed by histopathology were retrospectively analyzed.According to the median overall survival(OS)of 12 months,the patients were divided into the short OS group(53 cases)and the long OS group(72 cases),and were randomly divided into the training set(87 cases)and validation set(38 cases)in a ratio of 7:3.The radiomics features of each patient’s preoperative MRI sequences were extracted separately,including Contrast Enhanced-T1 Weighted Imaging(T1C),T1 Weighted Imaging(T1),T2 Weighted Imaging(T2)and T2 Weighted Imaging Fluid Attenuated Inversion Recovery Sequence(T2F).Stable radiomics features were selected by Intraclass Correlation Coefficients(ICCs),followed by t-test and the Least Absolute Shrinkage and Selection Operator Algorithm(LASSO)screening and dimensionality reduction of stable radiomics features.1.According to the selected features,establish three machine learning classification algorithm models to judge the ability of features to distinguish OS in GBM patients,including Random Forest Algorithm Model(RF),Support Vector Machine Algorithm Model(SVM)and Logistic Regression Algorithm Model(LR).The Receiver Operating Characteristic Curve(ROC)and Area Under the Curve(AUC)were used to evaluate the predictive performance of these models.2.According to the selected features and their corresponding weights,the Radiomics Score(Radscore)of each patient was calculated separately.The independent predictors were found through univariate and multivariate logistic regression based on the Radsocres and basic clinical features,including age at diagnosis,gender,preoperative Epilepsy Status(pEPI),preoperative Karnofsky Performance Status Score(pKPS),tumor location(frontal,temporal,parietal,occipital,insula,corpus callosum),and cerebral hemisphere location(left,right,bilateral),then the radiomics nomogram beneficial to clinical application was constructed.The predictive performance of the nomogram was evaluated by ROC,calibration curve and discrimination,and the clinical application value of the nomogram was evaluated by Decision Curve Analysis(DCA).ResultsA total of 5216 radiomics features were extracted from the four preoperative MRI sequences included in each patient,5060 stable features were screened by ICCs,and 21 radiomics stable features were finally included.Among the three machine learning classification algorithm model prediction results,the SVM model performed the best,with AUC of 0.97 and 0.75 in the training cohort and validation cohort,respectively.In training cohort,validation cohort and total cohort,Radscore was statistically significant in between the short OS group and the long OS group(p<0.001).The univariate and multivariate Logistic regression results showed that the Radscore was statistically significant(OR=5941.499,95%CI(239.983-336406.47),p<0.001).The ROC showed that the nomogram exhibited good predictive performance,with AUCs of 0.877 and 0.919 in the training and validation cohorts,respectively.The DCA showed that the Radscore-based nomogram had good clinical utility.Conclusion1.Preoperative multiparametric MRI radiomics features can better differentiate the survival stratification of GBM patients treated with standardization.2.In the machine learning classification models based on preoperative multi-parameter MRI radiomic features,S VM performed better.3.Among the included MRI sequences,the radiomics features of T1C sequence contributed the most to the prediction of survival stratification in GBM patients treated with standardization.4.The Radscore score was an independent predictor of survival stratification in GBM patients treated with standardization.5.The nomogram based on preoperative multiparametric MRI radiomics showed good predictive performance and clinical application value in predicting the survival stratification of GBM patients treated with standardization.