Multimodal Analysis of Vasogenic Edema in Glioblastoma Patients for Radiotherapy Planning

International audienceGlioblastoma (GBM) is the most common type of primary brain tumor, which is characterized by an infiltrative growth pattern. In current practice, radiotherapy planning is primarily based upon T2 FLAIR MRI despite its known lack of specificity in the detection of tu-mor infiltration. While hyperintensity on T2 FLAIR is widely considered to represent infiltrative tumor, it may also be caused by the presence of vasogenic edema (VE), caused by a leakage of fluid into the brain parenchyma. Distinguishing VE from infiltrative tumor could have im-pact on improving radiotherapy planning. In this paper we study a data set of 17 GBM patients treated with anti-angiogenic therapy for which a fast decrease of T2 FLAIR hypersignal is observed, which indicates the resolution of VE. We investigate if multimodal MRI acquisitions in-cluding diffusion tensor imaging can distinguish between VE and tumor infiltration prior to therapy. Using a random forest classifier, we show that, in this study, morphological information based on the contrast en-hanced T1 image explains up to 75% of the extent of VE. The information from different imaging modalities did not significantly improve the clas-sification. We then show that delineating the VE prior to therapy can have substantial impact on radiotherapy target delineation, leading to smaller treatment volumes and reducing potentially harmful radiation dose to normal brain tissue