Artificial Intelligence-Based Imaging Transcoding System for Multiplex Screening of Viable Foodborne Pathogens

Multiplexdetection of viable foodborne pathogens is critical forfood safety and public health, yet current assays suffer trade-offsbetween cost, assay complexity, sensitivities, and the specificitybetween live and dead bacteria. We herein developed a sensing methodusing artificial intelligence transcoding (SMART) for rapid, sensitive,and multiplex profiling of foodborne pathogens. The assay utilizesthe programmable polystyrene (PS) microspheres to encode differentpathogens, inducing subsequent visible signals under conventionalmicroscopy that can be analyzed using a customized, artificial intelligence-computervision, which was trained to decode the intrinsic properties of PSmicrospheres to reveal the numbers and types of pathogens. Our approachenabled the rapid and simultaneous detection of multiple bacteriafrom egg samples of <10(2) CFU/mL without DNA amplificationand showed strong consistency with the standard microbiologic andgenotypic methods. We adopted our assay through phage-guided targetingto enable the discrimination between live and dead bacteria.

Automated summary

In this study, a sensing method called SMART was developed for rapid, sensitive, and multiplex profiling of foodborne pathogens using artificial intelligence transcoding. The method utilizes programmable polystyrene microspheres to encode different pathogens and generate visible signals under conventional microscopy. A customized artificial intelligence-computer vision was trained to decode the microspheres' intrinsic properties and reveal the numbers and types of pathogens. The assay enables the rapid and simultaneous detection of multiple bacteria without DNA amplification, and shows strong consistency with standard microbiologic and genotypic methods. Phage-guided targeting is adopted to discriminate between live and dead bacteria.