Hybrid Approach to Colony-Forming Unit Counting Problem Using Multi-Loss U-Net Reformulation.

Area of Science:

• Biotechnology
• Computer Science
• Food Safety

Background:

• Colony-Forming Unit (CFU) counting is critical in biomedical and food safety but lacks a universal solution.
• Current deep learning methods like U-Net require post-processing for counting, often leading to inaccuracies due to pixel-based segmentation.
• Existing approaches struggle with artifacts and precise localization of microbial colonies.

Purpose of the Study:

• To develop a novel hybrid approach for precise in vitro CFU counting.
• To improve the accuracy and robustness of deep learning-based CFU detection.
• To introduce a fully automated system for CFU counting with a user feedback loop.

Main Methods:

• A reformulated multi-loss U-Net incorporating an auxiliary loss term in the bottleneck layer.
• A novel post-processing Petri dish localization algorithm that includes the agar plate and bezel.
• Integration with a uniform Petri dish illumination system and a web application for automated processing and user feedback.

Main Results:

• The proposed hybrid approach consistently outperformed single-loss U-Net and other models (density maps, YOLOv6) by 1-3% in mean absolute and symmetric mean absolute percentage errors.
• The multi-loss U-Net reformulation provided an auxiliary signal for better CFU localization.
• The Petri dish localization algorithm further enhanced counting accuracy.

Conclusions:

• The novel hybrid CFU counting approach significantly improves precision and accuracy in microbial colony detection.
• The multi-loss U-Net and localization algorithm effectively address limitations of existing segmentation-based methods.
• The fully automated system offers a robust and adaptable solution for in vitro CFU counting challenges.