Abstract
This paper proposes a novel cascaded U-Net for brain tumor segmentation.Inspired by the distinct hierarchical structure of brain tumor, we design acascaded deep network framework, in which the whole tumor is segmented firstlyand then the tumor internal substructures are further segmented. Consideringthat the increase of the network depth brought by cascade structures leads to aloss of accurate localization information in deeper layers, we construct manyskip connections to link features at the same resolution and transmit detailedinformation from shallow layers to the deeper layers. Then we present a lossweighted sampling (LWS) scheme to eliminate the issue of imbalanced data duringtraining the network. Experimental results on BraTS 2017 data show that ourarchitecture framework outperforms the state-of-the-art segmentationalgorithms, especially in terms of segmentation sensitivity.