DA-UNet: Deformable Attention U-Net for Nucleus Segmentation

Abstract: Cell nucleus segmentation plays a significant role in Computer-Aided systems for cancer diagnosis. However, complex visual features, such as blurring and irregular shapes, increase the difficulty of segmentation. This paper proposes a deformable attention U-Net (DA-UNet) to enhance the learning of nucleus complex visual features. Based on the traditional U-Net, we introduce a deformable attention (DA) module, which aims to learn the more suitable shape features by the attention mechanism and deformable convolution. Experiments on the 2018 Data Science Bowl and MoNuSeg datasets show that the proposed DA-UNet can achieve good results.