Radiomics For Survival Analysis And Prediction In Glioblastoma (Gbm)-A Preliminary Study

To evaluate the value of radiomics for survival analysis and prediction of glioblastoma (GBM) patients treated with chemoradiation therapy (CRT). Twenty-eight newly diagnosed GBM patients who received CRT at our institution between 2008 and 2014 were retrospectively studied. In addition to demographics (age, race, gender) and clinical parameters (KPS, extent of resection, radiation therapy type, dose and status, whether concurrent temozolomide was adjusted/stopped), 255 imaging features were extracted from 3 gadolinium-enhanced T1 weighted MRIs for 2 regions of interest (ROIs) (the surgical cavity and its surrounding enhancement rim). The 3 MRIs were at post-operation, 1-month and 3-month post-CRT. For the radiomics approach, imaging features comprehensively quantified the intensity, spatial variation (texture), geometric property, and their spatial-temporal changes for the 2 ROIs. Overall survival (OS) and progression-free survival (PFS) were analyzed using univariate and multivariate analysis. Machine learning models (logistic regression (LR), support vector machine (SVM), decision tree (DT), neural network (NN)) were applied to evaluate the survival prediction capability of algorithmically selected features. The number of cases and percentage of cases predicted correctly were collected and AUC (area under the receiver operating characteristic (ROC) curve) were determined after leave-one-out cross-validation. For univariate survival analysis, 28 features (1 demographic, 2 clinical and 25 imaging) were statistically significant (P<0.05) for both OS and PFS. The types of imaging features were volumetric, intensity, geometry, texture and their spatial-temporal changes. For multivariate survival analysis, 14 of the features remained statistically significant (P<0.05) for OS and 20 features remained statistically significant (P<0.01) for PFS. When all the features were used by machine learning models to predict the survival, 24 features were algorithmically selected. High prediction accuracy of OS was achieved by using NN (96%, 27 of 28 cases were correctly predicted, AUC = 0.99), LR (93%, 26 of 28 cases were correctly predicted, AUC = 0.95) and SVM (93%, 26 of 28 cases were correctly predicted, AUC = 0.90). When predicting PFS, NN obtained the highest prediction accuracy (89%, 25 of 28 cases were correctly predicted, AUC = 0.92). In this preliminary study, radiomics approach combined with patients’ demographics and clinical parameters are promising (even MGMT status was not available) for survival analysis and prediction in GBM patients treated with CRT. To achieve more accurate predictions, advanced machine learning models should be considered.