Using Convolutional Neural Networks to Segment Images of Packed Epithelial Cell Layers for Inferring Cellular Forces

Cellular force inference is a powerful tool that provides insights into the growth and morphogenesis of epithelial cell layers. To infer intercellular tensions and pressures between cells, it is mandatory to first segment a microscope image of the packed epithelial cell layer with its cell borders marked, for example by using a fluorescent protein such as E-cadherin-GFP. Current procedures depend on guided watershed algorithms. Here we present, CS-Net, a novel convolutional neural network based on a modified U-Net structure to transform a raw microscope image to a mask of pixels defining cell boundaries and cell interiors. Our neural network has an accuracy of 90% with an AUC-ROC of 0.86. CS-Net is made accessible through an accessible web API, and it is designed to be used in conjunction with a Python-based cellular force inference tool.