Poorly Differentiated Colorectal Gland Segmentation Approach Based on Internal and External Stress in Histology Images

Histological assessment of glands is one of the key factors in colon cancer grading. We propose an approach for segmentation of glands in colon cancer histopathology images, which bases on the characteristics of lumens and rough gland boundary. First, we present a U-net for stain separation of H& E image to obtain the H-stain, E-stain, and background stain intensity maps. Subsequently, the epithelial nuclei was identified on the H stain intensity map, and the lumen segmentation was performed on the background intensity map. Then, we employ the axis of least inertia and chain to establish a connection between the lumen and the epithelial nuclei. In the light of connection of the lumen and the epithelial nuclei, more accurately gland segmentation is performed based on the rough gland boundary. The proposed stain separation approach is unsupervised, and the stain separation makes the information contained in the H& E image more independent and easy to identify, so as to deal with the uneven stain intensity and the inconspicuous stain difference. We compare our results on a publicly available data set. The results show that the segmentation approach combining the characteristics of lumens and rough gland boundary has improved the segmentation accuracy.