Yeast cell segmentation in microstructured environments with deep learning.

Area of Science:

• Microscopy and quantitative biology
• Computational biology and image analysis
• Yeast cell biology

Background:

• Accurate cell segmentation is crucial for single-cell analysis from microscopy data, especially in complex microstructured environments.
• Existing tools for yeast in microstructures often rely on traditional machine learning, limiting performance.
• Deep learning offers potential for advanced image segmentation tasks.

Purpose of the Study:

• To develop and demonstrate deep learning models for multiclass segmentation of yeast cells within microstructured environments.
• To differentiate yeast cells from similar microstructures using convolutional neural networks.
• To provide accurate and fast segmentation solutions for systems and synthetic biology applications.

Main Methods:

• Implementation of U-Net based semantic segmentation.
• Application of Mask R-CNN for direct instance segmentation.
• Training, validation, and testing using curated datasets of yeast in microstructures.

Main Results:

• Convolutional neural networks achieved robust segmentation of yeast cells in microstructured settings.
• The developed models outperformed previous state-of-the-art methods in both segmentation accuracy and processing speed.
• Demonstrated utility in a typical systems or synthetic biology use-case.

Conclusions:

• Deep learning models, specifically U-Net and Mask R-CNN, provide effective solutions for yeast cell segmentation in microstructures.
• The enhanced accuracy and speed enable advanced applications like online monitoring and closed-loop experimental design.
• The study offers accessible code and data for reproducible research in quantitative yeast cell biology.