High-Speed, Label-Free Antimicrobial Susceptibility Testing in Picoliter Droplets: Combining Cage-Based Phase Contrast Microscopy with YOLO-SAM Algorithm

Rapid rise in antimicrobial-resistant bacteria highlights the urgent need for interventions to combat antimicrobial resistance (AMR). Antimicrobial susceptibility testing (AST) is crucial for antibiotic discovery and efficacy testing. However, existing AST systems are costly and labor-intensive. Herein, we developed an automatic, label-free, and high-throughput AST system called Cage-Based Phase Contrast Microscopy with You Only Look Once-Segment Anything Model (YOLO-SAM) Algorithm Droplets AST (CYSDrop AST) for antibacterial drug analysis. When using this system, the initial bacterial concentration should be controlled at around OD600 = 0.05 to maximize the encapsulation of one bacterium per droplet, and facilitate the subsequent statistical analysis. Because of low-cost T-junction microfluidic chips that encapsulate bacteria at single-cell level into picoliter droplets, the per-assay cost is reduced to at least one-third of that of traditional methods. The total sample-to-result time is shortened to 4 h, which includes a 2 h bacterial cultivation step. We used colistin susceptibility testing as a model system. The system determined the colistin susceptibility of E. coli strains within 4 h and achieved 100% categorical agreement with the reference broth microdilution (BMD) method. The core advantage of this method lies in its quantitative analysis of bacterial load, which achieves high consistency (89.6%) with manual annotation, complemented by its near-perfect performance in discriminating between positive and negative droplets (AUC = 0.994, AP = 0.988). Furthermore, we demonstrated its utility by testing colistin resistance in urine samples. This system is particularly valuable for guiding the use of last-line antibiotics and optimizing therapeutic strategies for patients with multidrug-resistant infections.