LIVECell

LIVECell provides a large, expert-annotated dataset of phase-contrast live-cell images for training and benchmarking CNN-based deep learning models for label-free cell segmentation.

Key Features:

• Large-Scale Dataset: Contains over 1.6 million manually annotated and expert-validated phase-contrast images covering diverse cell morphologies and culture densities.
• Manual Annotation and Expert Validation: Annotations were curated and validated by experts to provide high-quality ground truth for segmentation tasks.
• Deep Learning Integration: Intended for training and evaluating convolutional neural network (CNN)-based and other deep learning segmentation models.
• Benchmark Suite: Includes a proposed benchmark suite to comprehensively assess segmentation accuracy and model performance.
• Challenging Imaging Scenarios: Provides annotated examples representative of label-free phase-contrast imaging, including low-contrast and high object density conditions.

Scientific Applications:

• High-throughput quantitative imaging: Supports quantitative analysis workflows for large-scale imaging experiments.
• Two-dimensional cell culture studies: Enables segmentation and analysis of cells in 2D culture conditions using phase-contrast microscopy.
• Machine learning model training: Facilitates training of machine learning models to automate label-free cell segmentation when fluorescent labels are unavailable.
• Cellular behavior and morphology analysis: Supports investigation of cellular behavior, morphology, and cell–cell interactions via single-cell segmentation from phase-contrast images.

Methodology:

Uses phase-contrast microscopy images with expert manual annotation and supports training and evaluation of CNN-based deep learning models using the provided benchmark suite to assess segmentation accuracy.