Histopathology images are used to assess the status of certain biological structures and to diagnose diseases such as cancer. In computer-assisted diagnosis (CAD), nuclear segmentation for histopathology images is an essential prerequisite. In recent years, deep-learning technology has been gaining popularity in the field of nuclei segmentation. However, nuclei segmentation is still faced with challenging a lot of difficulties due to (1) staining intensity inhomogeneity, (2) background noise caused by preprocessing, (3) blurred boundaries due to a large number of overlapping cells. Furthermore, in histopathology imaging, the number of data samples in the dataset is relatively low, preventing deep convolutional neural networks (CNNs) from segmenting nuclei images with high accuracy like in other vision applications. To overcome the above difficulties, we propose a two-stage deep learning network for nuclei segmentation tasks. It is the first stage network responsible for coarse segmentation, and the second stage network for refined segmentation. In comparison with traditional network architectures, our method achieves near SOTA performance in the nuclei segmentation task.