Gastrointestinal cancer is a global health threat. Pathology image analysis is crucial for its diagnosis, and deep learning based methods have shown potential in improving diagnostic efficiency. This paper presents E-MRNet, an edge-morphological refinement network with multi-scale feature enhancement for medical pathology image segmentation. The proposed network, based on MMUNet, integrates convolutional neural networks and morphological methods. It uses Multi-Scale Convolutional Modules (MCNB) and Double - Attention Multi-Scale Convolutional Modules (Datt-MCNB) in the encoder and decoder, and incorporates an Erosion-Dilation Module (EDM) in skip connections and an Enhanced Edge Feature Module (EEFM) for edge refinement. Experiments on an in-house dataset from Shanghai Ruijin Hospital and the public MoNuSeg dataset demonstrate that E-MRNet outperforms several competitive networks such as U-Net, nnUNet, and UCTransNet. On the in-house dataset, E-MRNet achieves a Dice score of 72.70% and an IoU of 57.53%, and on the MoNuSeg dataset, it reaches a Dice score of 82.38% and an IoU of 70.24%. This indicates that E-MRNet can effectively segment cell nuclei in pathology images, providing a more accurate and efficient approach for gastrointestinal cancer diagnosis.