A Novel Antibacterial Susceptibility Test From Microscopic Images by Deep Learning

Antibacterial susceptibility testing is fundamental for evaluating the efficacy of antimicrobial agents and addressing the growing challenge of bacterial resistance. Traditional methods, such as the Kirby-Bauer disk-diffusion test, are conducted at the macroscopic level, which restricts accuracy and disregards essential microscopic bacterial dynamics. In this study, we introduce a novel methodology that integrates microscopic imaging with deep learning to measure inhibition zones of Staphylococcus aureus, providing enhanced precision and deeper insights into bacterial behavior. A YOLOv8-based neural network was optimized using 74 microscopic training images and validated with 78 independent cases. The proposed approach achieved an accuracy of 89.74% within a tolerance of ±0.2 mm, exhibiting a strong Pearson correlation with expert measurements (r = 0.94). This performance represents a 90% reduction in measurement error compared to conventional Petri dish evaluations, where variability often reaches 2.0 mm. Additionally, the system demonstrated robustness across different microscope models, including one not employed during training, highlighting its generalization capacity. Beyond accuracy, this methodology enables automated, reproducible, and rapid measurements, significantly reducing analysis time while eliminating operator subjectivity. Importantly, by capturing bacterial growth dynamics at the microscopic scale, the method allows early detection of bacterial susceptibility, potentially reducing clinical decision-making times from days to hours. This advancement holds considerable promise for antimicrobial nanomaterial evaluation, epidemiological studies, and clinical diagnostics. To encourage transparency and facilitate further research, all datasets, codes, and experimental setups developed in this work have been made publicly available, ensuring reproducibility and promoting the extension of this methodology to other bacterial species.