Methods for identification of salmonella in a broiler chicken carcass rinse with hyperspectral microscope images
Eady, Matthew Brent
MetadataShow full item record
Salmonella is a pathogenic bacterium of concern to the poultry industry. Traditional methods for detecting Salmonella from broiler carcass rinses require time consuming or costly procedures. Hyperspectral microscope images (HMI) apply spectral imaging to identify bacterial cells for a potential early and rapid detection method. Spectral imaging can be used to detect many color bands, resulting in a unique spectral profile of an object. The objective was to use darkfield HMI to identify Salmonella cells from a broiler carcass rinse. Four studies were conducted to develop a better understanding of the methodology, followed by application to the carcass rinse. First, we compared metal halide lamps and tungsten halogen lamps. Both lighting systems classified five Salmonella serotypes at 100% accuracies. The halogen signal could not be reduced to fewer bands, whereas the halide could be reduced to peaks at 546 and 590 nm. Second, calibration was performed with wavelength and radiometric lamps, determining the HMI system was optimal less than 300 ms and 3.5% gain, with signal-to-noise ratios approaching equilibrium at higher settings. Third, environmental stress factors of incubation temperature and pH were applied to the same strains of Salmonella, in addition to incubation on four agar types. No significant difference was found with agars or incubation temperatures 27 – 42 °C (p > 0.05), while acidic pHs of 4.7 and 5.3 differed significantly from cells grown at pH 7.3 (p < 0.01). Fourth, an experiment developed an unsupervised classification approach. Four bacteria were combined into one image, with a stepwise linear discriminant analysis (LDA) classifying mixed HMI at 98.1% accuracy. Finally, broiler chicken carcass rinses were centrifuged without enrichment or incubation for early detection. Salmonella was not initially present in the carcass rinse. A second treatment containing Salmonella spiked rinse was used. Stepwise LDA was applied for unsupervised classification of cells. Salmonella was found at a high level in the spiked treatment, while other organisms in the rinse were noted. This project addresses several fundamental questions of the methodology, and shows that identification of the bacterial cells is possible with HMI collected from a broiler chicken carcass rinse.