MC-Net: An Endometrial Cancer Pathological Multy Class Image Segmentation Method

Pathological image segmentation is a widely used technique for classifying lesions and surrounding tissue types. Developing high-precision segmentation algorithms relies heavily on a large dataset of accurately labeled digital pathology images; however, the labeling process is both time-intensive and laborious, and such well-labeled images are scarce. As a result, pathology image segmentation presents a significant challenge. In this paper, we propose an MC-Net-based framework for multi-class segmentation of endometrial cancer pathology images. The framework utilizes a semi-supervised segmentation approach, making it adaptable to datasets with varying annotation volumes and diverse application scenarios. By leveraging multi-class transformers to learn the irregular shapes of pathological tissues, MC-Net achieves high segmentation accuracy. Experimental results demonstrate that, compared to existing segmentation methods, MC-Net effectively performs multi-class segmentation and diagnostic classification of pathological tissues, thereby providing valuable technical support for the pathological diagnosis of endometrial cancer and enabling accurate staging of the disease through automated analysis.