Deep learning convolutional neural networks accurately classify genetic mutations and survival in gliomas

Gliomas represents a heterogeneous group of tumors with variable response to therapy despite sharing overlapping morphologic features. These differing outcomes partly relates to the multiple genetic mutations. For example, mutations in isocitrate dehydrogenase (IDH1) demonstrate significantly better survival compared to their wild counterparts^1,2. Therefore, an obstacle in glioma imaging analysis is that radiographic interpretation fails to account for the tumoral genetic variance, making it difficult to integrate clinically relevant biological activities. The primary objective of this abstract is to use a convolutional neural network (CNN) approach to discover specific imaging patterns predictive of the underlying genetic alterations of gliomas.