Abstract: To address the issue of limited segmentation accuracy caused by the complex structure of pathological images, the blurred boundaries between lesion areas and normal tissues, as well as the low contrast, a dual-branch fusion model PFTransCNN (Parallel Fusion Transformer and CNN) based on convolutional neural networks (CNN) and Transformer is proposed. This model aims to fully utilize the spatial and channel correlations in pathological images to achieve precise segmentation of cancerous tissues with blurred and smooth boundaries. Specifically, the model uses ResNet34 as the backbone network for the CNN branch and combines the Transformer module to extract multi-level features, capturing local correlations and remote dependency information. Through the Fusion module, the features of the two branches are interactively fused, enhancing semantic dependency relationships and effectively avoiding the loss of boundary features. Additionally, the Upsample Feature Modulator (UFM) is used to process the feature information in the upsampled branch, successfully capturing low-level spatial features and high-level semantic information, thereby achieving precise segmentation results. Experimental results on the GlaS, SEED, and MoNuSeg datasets show that the Dice coefficient of the proposed model reaches 91.61%, 90.32%, and 81.37% respectively, significantly outperforming existing methods, verifying its effectiveness and generalization ability in complex pathological image segmentation tasks.