Less is more

Mengran Wu, Zhiyang Liu

Keywords: Kidney segmentation, deep learning, contrast attention

Abstract: This paper focuses on developing a deep learning method for kidney and the kidney tumors and cysts. To further fuse the semantic information and the spatial information, contrast attention is added at the skip connections of a classical U-Net. To make a lighter model, we use a shallower U-Net with 4 times down strides. Evaluation results indicates that the proposed method is able to achieve a comparable performance with nnU-Net while using much fewer parameters.