CellSium

CellSium generates realistic image sequences of bacterial microcolonies grown in monolayers to produce synthetic ground-truth datasets for training deep learning-based segmentation models in microfluidic live-cell imaging.

Key Features:

• Realistic Image Synthesis: Produces realistic image sequences and time-lapse videos of bacterial microcolonies configurable to include or exclude fluorescence.
• Programmable Growth Models: Supports programmable cell growth models to customize cell-level behaviors and colony growth parameters.
• 3D Colony Geometries: Provides simulation-ready 3D colony geometries compatible with computational fluid dynamics (CFD) analyses.
• Ground-Truth Dataset Generation: Generates annotated synthetic datasets suitable for training deep learning-based segmentation models and neural networks.
• Microfluidic Context Configuration: Configurable to simulate scenarios relevant to microfluidic live-cell imaging of bacterial monolayers.
• Implementation: Implemented in Python.

Scientific Applications:

• Segmentation Model Training: Creation of large ground-truth datasets for training and evaluating deep learning-based segmentation models in live-cell imaging.
• Bacterial Colony Dynamics: Study of growth patterns and interactions in bacterial microcolonies grown as monolayers.
• Microfluidic Experiment Design: Simulation of microfluidic live-cell imaging scenarios to support experimental planning and analysis.
• Computational Fluid Dynamics Integration: Use of 3D colony geometries in CFD analyses to investigate microcolony behavior in fluidic environments.
• Bioinformatics and Computational Biology: Development and benchmarking of analytical tools and pipelines in microbiology, bioinformatics, and computational biology.

Methodology:

Implemented in Python, CellSium synthesizes realistic time-lapse image sequences (optionally fluorescent) of bacterial microcolonies using programmable cell growth models, produces simulation-ready 3D colony geometries compatible with CFD, and outputs ground-truth annotations for deep learning segmentation.