DeepSea: An efficient deep learning model for single-cell segmentation and tracking of time-lapse microscopy images

Abstract

Abstract Dynamics and non-genetic heterogeneity are two fundamental characteristics of basic processes of life such as cell division or differentiation. Time-lapse microscopy is the only method that can directly capture the dynamics and heterogeneity of fundamental cellular processes at the singlecell level with high temporal resolution. Successful application of single-cell time-lapse microscopy requires automated segmentation and tracking of hundreds of individual cells over several time points. Recently, deep learning models have ushered in a new era in the quantitative analysis of microscopy images. However, integrated segmentation and tracking of single cells remain challenges for the analysis of time-lapse microscopy images. This work presents a versatile and trainable deep-learning software, termed DeepSea, that allows for both segmentation and tracking of single cells in sequences of phase-contrast live microscopy images. Our segmentation model can easily be trained to segment phase-contrast images of different cell types with higher precision than existing models. Our tracking model allows for quantification of dynamics of several cell biological features of individual cells, such as cell division cycle, mitosis, cell morphology, and cell size, with high precision using phase-contrast images. We showcase the application of DeepSea by analyzing cell size regulation in embryonic stem cells. Our findings show that embryonic stem cells exhibit cell size control in the G1 phase of the cell cycle despite their unusual fast division cycle. Our training dataset, user-friendly software, and code are available here https://deepseas.org .