
Microbial Detection
Description
Foodborne pathogens such as E. coli, Listeria and Salmonella fall under zero-tolerance regulations, so fast and affordable detection is critical for food safety. Our prior work showed AI-assisted species-specific detection from white-light images of bacterial microcolonies within three hours, and built a data repository with more than 10 bacterial and 7 yeast species. This project develops the first multimodal AI models for microbial detection. They combine low-cost white-light microcolony imaging with headspace volatile profiles from an electronic nose (E-nose), which can easily be used in an industry QA/QC lab. Starting from pretrained vision and time-series models, we will align both modalities in a shared representation with self-supervised and contrastive learning, and train the model to forecast volatile dynamics for early detection. We will then fine-tune it to detect diverse microbes across different imaging and sensing devices, and to distinguish pathogenic from non-pathogenic species of the same genus. The result will be the first publicly available multimodal microbial model, and a step toward a foundation model for microbial phenotyping that also applies to plant and animal pathogens.
Team

Nitin Nitin
Principal Investigator

Xin Liu
Co Principal Investigator