Automated segmentation of glands to facilitate quantitative analysis in CD138 whole slide images using a KNet deep learning framework

Segmentation of glands in immunohistochemical (IHC) images is the top priority for automated evaluation of CD138 positive cells, which could exclude irrelevant regions and improve accuracy and effectiveness. In this paper we propose a novel patch-based pipeline with an integrated KNet-like deep learning framework to perform automated gland segmentation for CD138 whole slide images. The patch-based pipeline is composed of patch decomposition, tissue patch extraction, gland segmentation, patch merging, and gland mask padding. The KNet deep learning framework introduces SwinTransformer to extract multiscale patch features, and integrate Uper Net as basic semantic kernels into KNet framework to refine the gland segmentation results. The integrated framework is trained in an end-to-end way with a weighted cross-entropy loss and dice loss. The experimental results show that our proposed framework achieves state-of-the-art performance, and could produce accurate gland masks for real whole slide images, which lays a solid foundation for automated quantitative analysis of CD138 images.