Contrastive learning for cell division detection and tracking in live cell imaging data

Area of Science:

• Cellular dynamics and live-cell imaging
• Biotechnology and bioimaging

Background:

• Fluorescent live-cell microscopy is crucial for studying cellular processes but limited by photo-toxicity.
• Low temporal resolution compromises cell tracking and division event detection, hindering dynamic process studies.

Purpose of the Study:

• To develop an integrated methodology for improved cell division detection and tracking in low temporal resolution microscopy.
• To enhance the analysis of cellular dynamics while maintaining cell viability.

Main Methods:

• Utilized contrastive learning to generate robust cell representations from time-based augmentations.
• Developed a graph optimization method for cell track identification using learned representations and division events.

Main Results:

• Achieved significant performance gains in cell division detection and tracking accuracy.
• Demonstrated effectiveness across both native and reduced temporal resolutions on diverse datasets.

Conclusions:

• The methodology enhances adaptability to varying temporal resolutions for precise live-cell microscopy data analysis.
• Supports extended observation periods for drug screening and biological studies by preserving cell viability.
• Facilitates deeper insights into cellular mechanisms and potential therapeutic research advancements.