Fig. 1: UNSEG framework.
From: UNSEG: unsupervised segmentation of cells and their nuclei in complex tissue samples
Input is a two-channel image comprising of nucleus (channel 1) and cell membrane (channel 2) marker expressions. a A priori spatial probability distributions of nucleus and cell membrane marker expressions. b Likelihood map of a pixel to belong to the nucleus or cell membrane, quantified through the visual contrast function, which mimics human perception. c A posteriori local and global semantic segmentation masks respectively capturing local morphological heterogeneity and global nucleus and cell membrane topology. d Instance segmentation of nuclei from semantic segmentation masks. e Instance segmentation of cells based on individual segmented nuclei and semantic masks. Nucleus and cell segmentation results of (d, e) form the UNSEG output. See “Methods” for more details.
