Predicting Spatial Patterns of Recurrence for Glioblastoma Using Multi-Parametric MRI Classification

Vanessa H. Clark¹, Michel Bilello¹, Priyanka Bhatt¹, Xiao Da¹, Elias Melhem¹, Arastoo Vossough¹, Ron Wolf¹, Christos Davatzikos¹, Ragini Verma¹

We aim to provide a spatial map of predicted tumor recurrence in subjects with glioblastoma which can be used to guide treatment towards improving quality of life and survival. Using 14 subjects and 9 imaging modalities, we use support vector machine classification to generate a probability map representing tissue with imaging characteristics similar to recurring/non-recurring tissues. With average 97% specificity, 27% sensitivity, and 0.79 AUC, these probability maps show potential for multimodal imaging and classification to predict patterns of tissues likely to recur that are not obvious to the human eye, including prediction of tumor recurrence at specific time intervals.