P04.27 Texture feature analysis to predict overall survival in recurrent glioblastoma treated in the DIRECTOR trial

Abstract

INTRODUCTION: Recently, quantitative radiomics approaches have shown their potential to predict clinical outcome and O6‐methylguanine DNA methyltransferase (MGMT) promoter methylation status in patients with newly diagnosed glioblastoma. However, for recurrent glioblastoma scarce data on the value of quantitative radiomics is available. This study was designed to evaluate the association of clinical outcome with radiomic features in patients with recurrent glioblastoma. MATERIALS AND METHODS: We analyzed imaging data of the DIRECTOR trial, a prospective randomized multicenter study comparing two dose‐intensified temozolomide regimens in recurrent glioblastoma (N=105); data from 54 patients with complete imaging data either with or without repeat surgery were available for image analysis. The contouring of recurrent tumor was performed on gadolinium‐enhanced T1‐weighted images (T1+Gad MRI). We used a newly developed in‐house software for radiomic feature extraction and calculation (43 features describing tumor shape and tumor texture). Radiomic features were first preselected in the principal component analysis and later used in the multivariable analyses. A cox proportional hazard model was trained to predict progression free survival (PFS) and overall survival (OS) and a logistic regression was used to predict MGMT promoter methylation status. RESULTS: Five features correlated with PFS (P < 0.05) and three features correlated with OS (P < 0.05) in the univariable Cox regression analysis of patients with recurrent glioblastoma. In contrast, no parameter correlated with MGMT promoter methylation status (P > 0.05). Upon multivariable analysis, an association between two partially overlapping features and PFS and OS was found (PFS features: sum of entropy and large size high gray‐level emphasis (lshge); concordance index (CI), 0.622; Wald test, P = 0.002; OS: features, surface and lshge; CI, 0.615; Wald test, P = 0.024). CONCLUSION: In our analysis, both PFS and OS, but not MGMT promoter methylation status, were associated with texture features on T1+Gad MRI in patients with recurrent glioblastoma treated with alkylating agent chemotherapy. This proof of principle investigation should be validated on an independent external dataset of recurrent glioblastoma patients in order to establish potential imaging biomarkers with predictive potential.