Automated Identification and Measurement of Antimicrobial Inhibition Zones Using YOLOv8n and Grad-CAM

The global rise of antimicrobial resistance poses a critical threat to public health, necessitating rapid, accurate, and standardized methods for antimicrobial susceptibility testing. In this study, we propose an artificial intelligence–driven framework for the automated detection, quantification, and interpretation of inhibition zones in disk diffusion assays using the YOLOv8n object detection model. A curated dataset of high-resolution Petri dish images containing antibiotic discs tested against Escherichia coli, Salmonella, and Staphylococcus aureus under standardized imaging conditions was used for model training and evaluation. The system automatically detects inhibition zones, measures their diameters in millimeters, and classifies bacterial susceptibility according to Clinical and Laboratory Standards Institute (CLSI) breakpoint criteria. Experimental results demonstrated robust performance, with a detection accuracy of 91.24%, precision of 90.86%, recall of 83.27%, and an F1-score of 86.45%. Model interpretability was enhanced through Gradient-weighted Class Activation Mapping (Grad-CAM), which verified that predictions consistently focused on biologically relevant regions. This end-to-end pipeline provides an objective, reproducible, and explainable approach to digital antibiogram analysis, with significant potential for integration into clinical microbiology workflows and large-scale antimicrobial resistance surveillance.