Weakly supervised cell instance segmentation under various conditions

Kazuya Nishimura, Chenyang Wang, Kazuhide Watanabe, Dai Fei Elmer Ker, Ryoma Bise

Abstract: Highlights•Propose a weakly-supervised cell instance segmentation method based on a convolutional neural network.•Use point-level annotations as weak label instead of using pixel-level cell boundary annotations.•Demonstrate robustness on various imaging conditions such as phase-contrast and DIC and fluorescent images.•Perform instance segmentation without any human annotation cost by using nuclei-stained cells.