Detection and segmentation of brain tumors from MRI using U-Nets
In this paper, we exploit a cascaded U-Net architecture to perform detection and segmentation of brain tumors (low- and high-grade gliomas) from magnetic resonance scans. First, we detect tumors in a binary-classification setting, and they later undergo multi-class segmentation. The total processing time of a single input volume amounts to around 15 s using a single GPU. The preliminary experiments over the BraTS’19 validation set revealed that our approach delivers high-quality tumor delineation and offers instant segmentation.