Learning Hierarchical Semantic Correspondence and Gland Instance Segmentation

摘要

The morphology statistics of colon glands is a key feature for pathologists to diagnose colorectal cancer. Current gland instance segmentation methods show good overall performances, but accurate segmentation of extremely deformed glands in highly malignant cases or some rare benign cases remains to be challenging . In this paper, we propose a hybrid model that learns hierarchical semantic feature matching from histological pairs in an attentive process, where both spatial details and morphological appearances can be well preserved and balanced, especially for the glands with severe deformation. A consistency loss function is also introduced to enforce simultaneous satisfaction of semantic correspondence and gland instance segmentation on the pixel-level. The novel proposed model is validated on two publicly available colon gland datasets GlaS and CRAG. The model successfully boosts the segmentation performances on greatly mutated or deformed cases, and outperforms the state-of-the-art approaches.