LIVECell-A large-scale dataset for label-free live cell segmentation

Area of Science:

• Cell Biology
• Bioimaging
• Machine Learning

Background:

• High-throughput quantitative imaging using light microscopy and 2D cell culture aids biological studies.
• Accurate cell segmentation is crucial for analyzing complex biological questions but challenging in low-contrast, high-density images.
• Deep learning excels at image segmentation but requires extensive annotated data, lacking for label-free cellular imaging.

Purpose of the Study:

• To introduce LIVECell, a comprehensive dataset for label-free cellular imaging.
• To facilitate the development and benchmarking of deep learning models for cell segmentation.

Main Methods:

• Creation of a large, manually annotated, and expert-validated dataset (LIVECell) of phase-contrast images.
• Dataset comprises over 1.6 million cells with diverse morphologies and culture densities.
• Training and evaluation of convolutional neural network (CNN) models using the LIVECell dataset and proposed benchmarks.

Main Results:

• LIVECell provides a high-quality resource for training deep learning models in label-free cell segmentation.
• Demonstrated the utility of LIVECell by training and evaluating CNN models.
• Established benchmarks for assessing segmentation accuracy.

Conclusions:

• LIVECell addresses the need for annotated data in label-free cellular imaging.
• The dataset and benchmarks will advance deep learning applications in cell image analysis.
• Facilitates high-throughput quantitative imaging and biological discovery.