Multicenter study to demonstrate radiomic texture features derived from MR perfusion images of pseudoprogression compared to true progression in glioblastoma patients.

Abstract

2016 Background: To differentiate between pseudoprogression and true progression in patients with glioblastoma using MR perfusion radiomic texture analysis (TA). Methods: 98 patients with pathologically-proven diagnosis of GBM were retrospectively included in this IRB approved HIPAA compliant study. All patients underwent DSC and DCE Perfusion MRI as part of their routine clinical care. Images were analyzed using Nordic ICE 2.3 (NordicNeuroLab) ; rCBV and ktrans maps were obtained. Subsequently, 3D slicer 4.3.1(http://www.slicer.org) was used to segment the entire tumor on the different processed maps to create a volume of interest (VOI) for Radiomic TA. Multiple invariant texture features where then extracted from each VOI. 475 invariant texture features were applied to each map. Leave-one-out cross-validation (LOOCV), receiver operating characteristic (ROC), Kaplan Meier, and multivariate Cox proportional hazards regression analyses were used to assess the relationship between texture feature and pseudoprogression and true progression. Results: Variance and sum entropy were the two most significant radiomic features that discriminated between pseudoprogression and true progression. P value, AUC, specificity and sensitivity were 0.03, 89.26%, 81.82%, and 100% respectively. Conclusions: Radiomic TA derived from perfusion images can be helpful in determining true versus pseudoprogression in GBM. Further, this study illustrates successful application of radiomic TA as an advanced processing step for different MRI perfusion maps (DCE, DSC).