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

The rising threat of antimicrobial resistance has underscored the urgent need for rapid, standardized, and automated antimicrobial susceptibility testing (AST) solutions. Leveraging recent advances in computer vision and deep learning, this study introduces a fully image-based diagnostic framework that integrates a YOLOv8n object detection model with a Convolutional Neural Network (CNN) to perform end-to-end bacterial identification and susceptibility classification directly from Petri dish images. The YOLOv8n model accurately localizes handwritten bacterial species labels with a mean Average Precision (mAP@0.50) exceeding 0.93, demonstrating robustness across diverse handwriting styles and imaging conditions. Complementarily, the CNN achieves a balanced accuracy of 94.7% and 100% sensitivity for susceptible cases by analyzing inhibition zone morphology without manual measurements or interpretive rules. Notably, the system achieved a 0% Very Major Error (VME) rate, ensuring no resistant isolates were misclassified as susceptible. The dual-model system generalizes effectively across experimental variations and produces high-confidence predictions, highlighting its potential to streamline AST workflows, minimize human variability, and enhance diagnostic reliability in clinical microbiology. • A fully image-based deep learning framework is proposed for automated antimicrobial susceptibility testing (AST). • YOLOv8n detects handwritten bacterial species labels with high accuracy ( mAP@0.50 > 0.93) across diverse imaging conditions. • CNN model predicts susceptibility categories (resistant/susceptible) with 94.7% balanced accuracy and 100% sensitivity for susceptible cases. • The integrated pipeline eliminates manual measurements and interpretive charts, enabling scalable, technician-independent AST workflows.