Article: LIVECell—A Large-Scale Dataset for Label-Free Live-Cell Segmentation
Light microscopy - combined with well-established protocols of two-dimensional cell culture - facilitates high-throughput quantitative imaging for the study of biological phenomena. Accurate segmentation of individual cells in images enables exploration of complex biological questions but can require sophisticated imaging processing pipelines. This is especially true in cases of low contrast and high object density. Deep learning-based methods are considered state-of-the-art for image segmentation but typically require vast amounts of annotated data. Until recently, there was no suitable resource available in the field of label-free cellular imaging.
In this study, we present LIVECell (Label-free In Vitro image Examples of Cells), a new dataset of manually annotated, label-free, phase-contrast images of 2D cell culture. LIVECell consists of more than 1.6 million annotated cells of eight morphologically distinct cell types (grown from early seeding to full confluence) and has undergone rigorous quality assurance to minimize bias in the annotations.
As a proof of concept of the use of LIVECell, we also present trained models developed to segment individual cells for application in new research to enable label-free, single-cell studies. Finally, in the interest of standardizing evaluation of such models, we propose a suite of benchmarks which will readily facilitate continued development and performance comparison of future models.
This article explores:
- Components of LIVECell - including benchmarks - to evaluate experiments and performance
- Details for assessing transferability between LIVECell and other datasets
- Discussion and review of the methodology behind LIVECell
Authors:
Christoffer Edlund, Timothy R. Jackson, Nabeel Khalid, Nicola Bevan, Timothy Dale, Andreas Dengel, Sheraz Ahmed, Johan Trygg, and Rickard Sjögren
