Rapid Bacterial Typing via Fluorescence Differential Staining and Deep Learning

Abstract Rapid and accurate bacterial typing is essential for the diagnosis, treatment, and surveillance of bacterial infectious diseases. Conventional approaches are time-consuming, labor-intensive, and prone to misidentification. Here, we present a rapid bacterial detection and classification strategy that integrates fluorescence differential staining with deep learning. Two fluorescent dyes, proflavine salts and 6-HEX, exhibiting markedly distinct binding affinities for Gram-negative and Gram-positive bacteria, were combined with Hoechst 33342 to establish a three-channel fluorescence fingerprinting system. A YOLOv8 deep-learning model was trained for automated bacterial recognition and classification, yielding an average identification accuracy of 87.7% across six clinically relevant pathogens, with accuracies reaching 94% for Staphylococcus aureus and 95% for Escherichia coli. The model also reliably distinguished species in mixed samples, with predicted relative abundances matching the inoculation ratios. This method offers high specificity, favorable quantitative capability, and rapid throughput, holding great promise for clinical pathogen diagnosis, food safety monitoring, and environmental microbiological detection.