Submitted to: Journal of Spectral Imaging
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/9/2016
Publication Date: 11/16/2016
Citation: Eady, M.B., Park, B. 2016. Rapid identification of Salmonella serotypes through hyperspectral microscopy with different lighting sources. Journal of Spectral Imaging. 5(a4):1-10.
Interpretive Summary: Hyperspectral microscopy is a form of optical detection that has previously been used as an early and rapid detection method for foodborne bacteria. These studies have shown promising results with a metal halide (MH) arc lighting source. The objective of this study is to compare the arc lamp to tungsten halogen (TH) lighting source, which has broad wavelengths compare to MH, assessing detection accuracies and robustness with different light sources. Hyperspectral microscope images (HMIs) of live foodborne bacterial cells from five Salmonella serotypes were collected with both lighting sources. It was found that key spectral bands in the visible ranges could be used for classification of the bacterial samples with MH. The experiments were repeated for validation of models with image subsets from images collected with MH and TH lighting sources. In this study, it was found that the TH, MH, and reduced MH samples had over 99% classification accuracy for the classification of five serotypes. The full MH and TH spectra yielded repeatable results for classifying foodborne pathogens.
Technical Abstract: Hyperspectral microscope imaging (HMI) has the potential to classify foodborne pathogenic bacteria at cell level by combining microscope images with a spectrophotometer. In this study, the spectra generated from HMIs of five live Salmonella serovars from two light sources, metal halide (MH) and tungsten halogen (TH) were compared to assess classification accuracy and robustness, between 450 – 800 nm. Ten key wavelengths between 594 – 630 nm were identified from MH HMIs, but TH band reduction decreased classification accuracy. Multivariate data analysis methods were applied with the principal component – linear discriminate analysis (PC-LDA) algorithm for the classification of the five Salmonella serotypes. PC regression was applied to the second and third repetitions to assess the repeatability of the experiment. PC-LDA classified serovar subsets (n = 1,800), reporting both MH and TH accuracies at 100%, while the reduced key MH bands achieved 99.4 – 100% accuracy. PC regression calculated the root mean squared error of cross–validation < 0.014 and a R2 > 0.948 for both full spectrum lights. MH or TH light source can be effectively used for discriminating bacteria HMIs on a serovar level, but TH bands cannot be reduced for the results.