Artificial Intelligence-Empowered Single-Cell Phenotyping for Rapid, Automated Pathogen Diagnostics

Accurate bacterial identification and antimicrobial susceptibility testing (AST) are essential for timely clinical decision-making. Conventional diagnostic methods typically require 24-48 h, leading to inappropriate empirical therapy, accelerating antimicrobial resistance, and worsening patient outcomes. Here, we present an integrated diagnostic platform that combines microfluidic single-cell bacterial detection with artificial intelligence (AI)-driven analysis for rapid phenotypic AST. The microfluidic device enables stable time-lapse detection of individual bacterial cells under designated, controlled antibiotic exposure, with a U-Net-based AI model that automatically segments and quantifies the drug response of individual bacterial cells using phase-contrast microscopy images. The trained model identified 96% of individual Escherichia coli (E. coli) cells and showed no false-positive predictions on bacteria-free images. Applied to AST, the integrated system quantified dose-dependent growth inhibition of E. coli exposed to ciprofloxacin and trimethoprim/sulfamethoxazole, yielding resistance profiles, along with minimum inhibitory concentrations (MICs), within 2 h. These results are consistent with standard broth microdilution, demonstrating the potential of integrating microfluidics and AI for rapid, automated, and single-cell-resolved pathogen diagnosis. Beyond rapid AST, the integrated microfluidic-AI platform was further evaluated for automated identification of targeted infections, including E. coli and Staphylococcus aureus (S. aureus) in the presence of blood matrices.