33、Morphology-Inspired Unsupervised Gland Segmentation via Selective Semantic Grouping

Morphology-Inspired Unsupervised Gland Segmentation via Selective Semantic Grouping

1. Introduction

Accurate gland segmentation from whole slide images (WSIs) is crucial for cancer diagnosis and prognosis. With the rise of deep learning, there has been growing interest in developing DL - based gland segmentation methods. However, these methods usually rely on large - scale annotated image datasets, which are costly and labor - intensive.

To reduce annotation costs, various methods have been explored, including weakly supervised semantic segmentation and annotation - free methods using conventional clustering and metric learning. But these methods have limitations, especially in cases of malignancy.

Unsupervised semantic segmentation (USS) met