I. Introduction
Cancer grading is of great clinical importance because it helps physicians to develop medical plans and evaluate the effectiveness of treatment [1]. Pathological images containing rich semantic information are widely used for cancer grading and diagnosis and are an important reference standard. Pathology image segmentation can highlight lesion areas to assist physicians in diagnosis and improve efficiency. However, pathology images with inconsistent staining, irregular morphology, and complex and diverse borders remain a challenging problem to achieve automatic and accurate segmentation.