Morphologically constrained and data informed cell segmentation of budding yeast.

Area of Science:

• Cell biology
• Bioimage analysis
• Microfluidics

Background:

• High-content image cytometry generates vast datasets, posing challenges for time-lapse experiment analysis.
• Existing automated single-cell segmentation methods struggle with microfluidic devices due to non-uniform fields of view and complex cell arrangements.
• Poor segmentation performance in microfluidics hinders long-term cell imaging and analysis.

Purpose of the Study:

• To develop a robust framework for cell tracking and segmentation in microfluidic devices.
• To leverage physical and temporal constraints for improved image analysis.
• To enhance the accuracy of automated cell segmentation in challenging imaging environments.

Main Methods:

• Introduced DISCO (Data Informed Segmentation of Cell Objects), a novel framework for cell segmentation and tracking.
• Integrated physical constraints of microfluidic traps, cell morphology, and temporal information (growth, motion).
• Utilized manually curated datasets for training and validation.

Main Results:

• DISCO demonstrated substantial improvements in both cell tracking and segmentation accuracy compared to existing software.
• The framework effectively handles the complexities of cell behavior within microfluidic environments.
• Performance gains were validated using curated ground-truth datasets.

Conclusions:

• DISCO offers a significant advancement in analyzing time-lapse imaging data from microfluidic cell cultures.
• The approach enhances the reliability of automated cell segmentation in non-ideal imaging conditions.
• The developed framework facilitates more accurate and efficient long-term cell studies.