Simultaneous brain tumor segmentation and molecular profiling using deep learning and T2w magnetic resonance images

Gliomas demonstrate diverse imaging features, variable response to therapy, and differences in prognosis. This is largely a function of genetic heterogeneity. Several key mutations serve as therapeutic and prognostic markers such as isocitrate dehydrogenase (IDH) mutation status, O6-methyl guanine-DNA methyltransferase (MGMT) promoter status, and 1p/19q co-deletion status. Currently, the gold standard for molecular marker determination requires tissue from either an invasive brain biopsy or surgical resection. Here we describe our work in developing highly accurate simultaneous deep learning segmentation and classification approaches for noninvasive profiling of molecular markers using T2-weighted magnetic resonance images only.