Submitted to: Journal of Spectral Imaging
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/3/2018
Publication Date: 3/6/2018
Citation: Eady, M.B., Park, B. 2018. Unsupervised classification of Salmonella Typhimurium from mixed bacteria cultures with hyperspectral microscope imaging. Journal of Spectral Imaging. DOI:10.1255/jsi.2018.a6.
Interpretive Summary: One in six people are affected by foodborne illness each year in the United States. Traditional detection methods are time consuming. Previously, hyperspectral microscope images (HMIs) was applied to detect pathogenic foodborne bacteria at the cellular level as a rapid and early detection method. In order to move the methodology towards more practical food industry use multiple species of bacteria need to be detected from the HMI we developed. Here, we tested mixtures of S. Typhimurium (ST) with three other species of common foodborne bacteria. The objective was to identify ST from the other three organisms. Hyperspectral microscope images containing all possible mixtures of ST with the other three bacteria were prepared. A cluster analysis statistical classification method was used to group similar cells together, based on their unique spectral signature. These clusters were verified through other statistical analyses including discriminant analysis and shape analyses. It was found that ST was classified at a selectivity of 97.9% and a selectivity of 97.5%. The shape metrics proved useful in identifying the cells as one of the four species. This project shows promise in identifying individual bacteria cells of different species from samples containing multiple species.
Technical Abstract: Salmonella is a leading cause of foodborne illness. Traditional detection methods require lengthy incubation periods or expensive reagent kits. Hyperspectral microscope images (HMIs) has been investigated as a means of early and rapid detection of bacteria by using a spectral signature that is unique to the organism. Previous use with bacteria has consisted of supervised classification of HMI collected for single culture HMIs. In order to move forward with HMI as a detection tool for the food industry, unsupervised classification of bacteria cells in mixed culture HMIs was investigated. S. Typhimurium (ST) was tested for unsupervised classification against combinations of three other foodborne pathogens E. coli (Ec), S. aureus (Sa), and L. innocua (Li). A divisive cluster analysis (CA) was implemented. CA cluster accuracy was checked by assigning a dummy variable of proposed CA classification, then carrying out a discriminant analysis (DA). Cluster profiling and species identification was done by calculating three shape metrics; the feret diameter ratio (FDR), circularity, and area of the cells. HMIs of ST with all possible combinations of the other three species were collected between 450 – 800 nm. From the mixed culture HMIs 700 bacteria cells were classified. Identification of ST cells showed a 0.9795 selectivity and a specificity of 0.9755. Shape metric mean values proved useful in determining the species identification of the unknown clusters from mixed culture HMIs. These results showed that it is possible to identify individual bacteria cells through HMIs containing more than one species.