Artificial-intelligence-driven segmentation and analysis of microbial cells

Area of Science:

• Microbiology
• Biotechnology
• Computational Biology

Background:

• Single-cell microbial analysis is crucial for physiology and biotechnology.
• Current manual annotation methods are inefficient and limit scalability.
• Automated quantitative analysis of cell growth and morphology is highly desirable.

Purpose of the Study:

• To develop an AI-driven system for automated microbial cell segmentation and quantitative feature analysis.
• To enable efficient monitoring of cell counts and morphological characteristics without manual annotation.
• To provide a scalable solution for high-throughput microbial cell analysis.

Main Methods:

• A four-component modular system: denoising, zero-shot segmentation (Segment Anything Model - SAM), post-processing, and feature extraction.
• Ablation experiments were conducted to evaluate individual component performance.
• The Segment Anything Model (SAM) with the SAM-H variant was optimized for microbial cell segmentation.

Main Results:

• Denoising and post-processing enhanced segmentation accuracy by 12.10% and 2.30%, respectively.
• The SAM-H model achieved a low average error rate of 3.0% on 1162 annotated *Escherichia coli* cells.
• The system successfully extracted morphometric and intensity features from *E. coli*, yeast, and cancer cell lines.

Conclusions:

• The developed framework automates quantitative analysis of microbial cells in high-resolution microscopy.
• This automation facilitates advanced research into microbial adaptations.
• The system has potential applications in studying extremophiles in challenging environments.