Nuclear detection and segmentation in pathological images is a crucial prerequisite for finding and forecast of numerous ailments. With the advancement of deep learning, high precision automatic segmentation of multi-organ nuclei pathological images is possible. The segmentation results require not only accurate foreground prediction but also careful annotation of every individual object. There are still several remaining challenges in multi-organ nuclei segmentation. Firstly, blurred boundaries and inconsistent staining make pixel-wise segmentation difficult to generate occlusive object masks. Secondly, the background noise will be retained or even enhanced after preprocessing. Thirdly, the differences in size, shape, and intensity of different organs make it harder to separate touching nuclei. Two novel weight maps of loss function are proposed, which make full use of structural information provided by raw image and its corresponding annotations, in order to supervise the network learning more available features. The Stain Refinement Weight Map focuses on the structural difference between Hematoxylin channel and Eosin channel, which has been neglected in existing methods, to highlight potential noise pixels. The Boundary-Enhancement Weight Map leverages the new boundary annotation of each individual object that has emerged in the recent dataset to help the network better divide cell clusters. We test our method on Multi-Organ Nuclei Segmentation Dataset and show that our proposed method has high accuracy in nuclei detection and can separate touching nuclei effectively in the segmentation task. Contribution-The main contribution of this paper is to propose two pre-defined novel weight maps to help the network extract better features.