Nuclei instance segmentation with dual contour-enhanced adversarial network

Abstract

The morphology of cancer cells is widely used by pathologists to grade stages of cancers. Accurate cancer cell segmentation is significant to obtain quantitative diagnosis. We proposed a dual contour-enhanced adversarial network to solve this challenge. The distance-transformed and contour-highlighted masks, and adversarial network are incorporated to improve individual cell segmentation capability. By evaluating quantitative individual cell segmentation results on 2017 MICCAI Digital Pathology Challenge, our method achieved best balance between precision and recall rate of individual cell segmentation compared to state-of-the-art cell segmentation methods.