A Deep-Learning Based System for Rapid Genus Identification of Pathogens under Hyperspectral Microscopic Images

Area of Science:

• Microbiology
• Artificial Intelligence
• Biotechnology

Background:

• Infectious diseases pose significant threats to human survival and incur substantial economic costs.
• Conventional bacterial identification methods are often slow, expensive, and labor-intensive.
• Rapid and automated identification of pathogenic bacteria is crucial for timely diagnosis and treatment.

Purpose of the Study:

• To develop and evaluate an AI-assisted system for automated, rapid bacteria genus identification.
• To combine hyperspectral microscopic technology with a deep-learning algorithm for enhanced bacterial analysis.
• To provide a user-friendly and efficient tool for clinicians and researchers.

Main Methods:

• Construction of an AI-assisted system integrating hyperspectral microscopy and a deep-learning algorithm (Buffer Net).
• Training and validation of the system using a custom dataset of over 130,000 hyperspectral images across 11 bacterial genera.
• Performance comparison against established deep learning models like 1D-CNN, 2D-CNN, and 3D-ResNet.

Main Results:

• The AI-assisted system achieved a high accuracy of 94.9% in bacteria genus identification.
• The developed system demonstrated superior performance compared to 1D-CNN, 2D-CNN, and 3D-ResNet.
• The system can identify pathogenic genera rapidly, with analysis taking approximately 30 seconds per hyperspectral microscopic image.

Conclusions:

• The AI-assisted system offers a rapid, accurate, cost-effective, and automated solution for bacteria genus identification at the single-cell level.
• This technology has the potential to significantly reduce identification time, potentially eliminating the need for bacterial cultivation.
• The user-friendly nature of the system makes it accessible for novices and beneficial for clinical applications.