Bridging Clinical Microbiology and Artificial Intelligence: An Image-Based Deep Learning Framework for Automated...

Area of Science:

• Medical Microbiology
• Computer Vision
• Artificial Intelligence

Background:

• Antimicrobial resistance (AMR) necessitates rapid, automated antimicrobial susceptibility testing (AST).
• Current AST methods can be slow, labor-intensive, and prone to human error.
• Advances in AI and computer vision offer potential for improved diagnostic workflows.

Purpose of the Study:

• To develop and validate a fully image-based diagnostic framework for bacterial identification and AST.
• To integrate YOLOv8n object detection and Convolutional Neural Network (CNN) for automated analysis of Petri dish images.
• To assess the system's accuracy, sensitivity, and error rates compared to traditional methods.

Main Methods:

• A dual-model deep learning framework combining YOLOv8n for label detection and CNN for inhibition zone analysis.
• Training and testing on diverse Petri dish images under various conditions.
• Evaluation of model performance using metrics such as mAP, balanced accuracy, sensitivity, and Very Major Error (VME) rate.

Main Results:

• YOLOv8n achieved mAP@0.50 > 0.93 for bacterial species label localization.
• CNN demonstrated 94.7% balanced accuracy and 100% sensitivity for susceptible cases.
• The system recorded a 0% VME rate, correctly identifying all susceptible isolates.

Conclusions:

• The image-based system offers a robust and reliable solution for automated AST.
• This framework has the potential to streamline clinical microbiology workflows and reduce diagnostic variability.
• High-confidence predictions from the dual-model system enhance diagnostic accuracy in identifying antimicrobial resistance.