A probabilistic approach to joint cell tracking and segmentation in high-throughput microscopy videos

Area of Science:

• Computational Biology
• Cell Biology
• Image Analysis

Background:

• High-throughput microscopy generates vast amounts of cell imaging data.
• Accurate cell segmentation and tracking are crucial for understanding cell behavior and lineage.
• Existing methods often rely on cell shape assumptions and struggle with complex datasets.

Purpose of the Study:

• To develop a general computational framework for analyzing high-throughput microscopy videos of living cells.
• To improve cell lineage construction by integrating cell segmentation and tracking.
• To overcome limitations of existing methods by not assuming cell shape.

Main Methods:

• A novel Bayesian inference framework for dynamic models is proposed.
• Time series analysis estimates temporal cell shape uncertainty and trajectory.
• A fast marching (FM) algorithm integrates inferred cell properties with image data for segmentation and association.

Main Results:

• The framework was tested on eight diverse time-lapse microscopy datasets.
• Promising results were achieved in detecting, segmenting, and associating planar cells.
• Performance surpassed the state of the art on the Fluo-C2DL-MSC dataset from the Cell Tracking Challenge.

Conclusions:

• The proposed computational framework offers a versatile approach for cell microscopy video analysis.
• It effectively handles spatial, temporal, and cross-sectional variations in data.
• The method demonstrates superior performance in cell detection, segmentation, and tracking, advancing the field.