Disc Diffusion Reader: an AI-powered potential solution to combat antibiotic resistance in developing countries.

Abstract

Introduction: Antimicrobial resistance (AMR) is a global health challenge, and antimicrobial susceptibility testing (AST) is vital for guiding treatment. Although widely used, the Kirby-Bauer method depends on skilled interpretation, which can be time-intensive and error-prone. This study explored the potential of an artificial intelligence (AI)-driven progressive web app (PWA) to automate the analysis of Kirby-Bauer test images, thereby enhancing accuracy and efficiency.

Methodology: Images of Kirby-Bauer test results were annotated to train the Faster R-CNN ResNet-50 to detect agar plates, inhibition zones, and antibiotic discs. MobileNetv2 was used for antibiotic disc classification. A Human-in-the-Loop (HITL) approach enabled technicians to correct errors and improve model performance through retraining. The PWA, built with VueJS and Python-PHP, provided real-time analysis aligned with the Clinical and Laboratory Standards Institute (CLSI) and the European Committee on Antimicrobial Susceptibility Testing (EUCAST) standards.

Results: The application achieved 92.95% accuracy for inhibition zone detection and 96.92% accuracy for antibiotic disc identification, with a performance improvement of 99.28% following HITL corrections. The measurements closely aligned with those of the technicians in 89.54% of the cases. The system processed up to 50 images per hour, supporting reliable and rapid AST workflow.

Conclusions: The AI-powered "Disc Diffusion Reader" demonstrated high accuracy and efficiency, by reducing interpretation variability in the AST workflows. Its scalability and adaptability, particularly in low-resource settings, make it a valuable tool for combating AMR. Continuous retraining and validation will ensure sustained reliability, and highlight the potential of AI-driven solutions in modern microbiology.