U²-Net and ResNet50-Based Automatic Pipeline for Bacterial Colony Counting

Area of Science:

• Microbiology
• Computer Vision
• Bioinformatics

Background:

• Accurate microbial colony counting is crucial for various laboratory applications.
• Manual counting is labor-intensive, prone to errors, and time-consuming.
• Automated systems can improve efficiency and reproducibility in microbial quantification.

Purpose of the Study:

• To develop an automated colony counting system using improved image preprocessing and convolutional neural network (CNN) assistance.
• To enhance the accuracy and automation of recognizing and counting single and multi-colony targets in laboratory cultures.
• To validate the system's performance in colony counting and adhesion classification.

Main Methods:

• An LED backlighting illumination platform was used for image acquisition of agar plate cultures.
• An image preprocessing algorithm based on light intensity correction was implemented for clearer colony-media differentiation.
• U²-Net models were employed for segmenting the Petri dish edge and the colony region.
• ResNet50 was utilized for the final automatic counting of segmented colony components.

Main Results:

• The U²-Net for Petri dish edge detection achieved an F1 score of 99.5% and MAE of 0.0033.
• The U²-Net for colony region segmentation achieved an F1 score of 96.5% and MAE of 0.005.
• The overall colony counting recovery rate was 97.82%, with excellent performance in adhesion classification.

Conclusions:

• The developed automated system demonstrates high accuracy and automation for microbial colony counting.
• The proposed pipeline, integrating light intensity correction, U²-Net segmentation, and ResNet50 counting, is a novel approach.
• This system offers a significant advancement for quantitative microbiology research and diagnostics.