CellSium: versatile cell simulator for microcolony ground truth generation.

Area of Science:

• Microfluidics
• Live-cell imaging
• Computational biology

Background:

• Deep learning models require large ground truth datasets for training.
• Microfluidic live-cell imaging presents unique challenges for data acquisition.
• Generating realistic synthetic data is crucial for advancing biological image analysis.

Purpose of the Study:

• To introduce CellSium, a novel cell simulator for synthesizing realistic image sequences of bacterial microcolonies.
• To provide a tool that addresses the need for large ground truth datasets in microfluidic live-cell imaging.
• To demonstrate the utility of simulated images for training neural networks.

Main Methods:

• CellSium is a flexibly configurable simulator developed in Python.
• It synthesizes realistic image sequences of bacterial microcolonies growing in monolayers.
• Supports synthetic time-lapse videos with and without fluorescence, programmable cell growth models, and 3D colony geometries.

Main Results:

• The simulated images generated by CellSium are suitable for training deep learning-based segmentation models.
• CellSium facilitates the creation of diverse synthetic datasets for various imaging conditions.
• The software supports advanced features like fluorescence simulation and integration with computational fluid dynamics.

Conclusions:

• CellSium effectively addresses the need for large ground truth datasets in microfluidic live-cell imaging.
• The synthetic data generated by CellSium can significantly aid in the development and training of neural networks for cell segmentation.
• CellSium is a valuable, open-source resource for researchers in computational biology and live-cell imaging.