Fully unsupervised symmetry-based mitosis detection in time-lapse cell microscopy.

Area of Science:

• Cell Biology
• Bioimaging
• Computational Biology

Background:

• Cell microscopy datasets exhibit high variability, complicating analysis.
• Supervised methods for mitosis detection require extensive manual annotations, which are challenging due to imbalanced cell populations.

Purpose of the Study:

• To develop a fully unsupervised framework for mitosis detection and mother-daughter cell association.
• To accommodate variations in cell appearance and dynamics in fluorescence microscopy data.

Main Methods:

• Utilizes daughter cell similarity to address symmetric cell divisions.
• Defines cell neighborhood using a stochastic Delaunay triangulation.
• Employs dynamic programming for optimization.

Main Results:

• Achieves promising mitosis detection results across diverse fluorescence microscopy datasets.
• Demonstrates effectiveness on both 2D and 3D sequences, including those from the Cell Tracking Challenge.
• Successfully performs mother-daughter cell association without manual supervision.

Conclusions:

• The unsupervised framework offers a robust solution for mitosis detection and cell lineage tracing.
• The method's adaptability to various datasets and imaging conditions makes it broadly applicable.
• Availability of code facilitates further research and development in cell division analysis.