|SEO, YOUNGWOOK - US Department Of Agriculture (USDA)|
|Hinton, Jr, Arthur|
Submitted to: Journal of Food Measurement and Characterization
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 12/14/2015
Publication Date: 1/7/2016
Citation: Seo, Y., Park, B., Hinton Jr, A., Yoon, S.C., Lawrence, K.C., Gamble, G.R. 2016. Identification of staphylococcus species with hyperspectral microscope imaging and classification algrorithms. Journal of Food Measurement and Characterization. 10(2):253-263.
Interpretive Summary: Recent foodborne outbreaks increase the threat to public health. The outbreak of food poisoning is caused by various foodborne pathogenic bacteria including Staphylococcus spp., Staphylococcus is widespread in the environment and has become one of the most commonly isolated pathogens in hospital-acquired infections. Due to the limitation of current methods in terms of speed, sensitivity and specificity, more rapid and accurate detection methods are needed for prevention of foodborne outbreaks for consumer protection. Although numerous detection methods for serotyping based on antibodies or genetic fingerprinting are being used with high accuracy, those methods are not cost-effective and often require intensive training fo consistent and accurate results. Therefore, development of rapid, simple and reliable techniques to detect and differentiate foodborne pathogenic bacteria serotypes and species are needed for food industry. Optical methods are good candidate to meet the requirement mentioned above. In this study, a hyperspectral microscopic imaging platform for acquiring spectral fingerprints from live bacterial cells, was used and classification methods were developed to identify foodborne pathogenic bacteria using hyperspectral microscope imaging.
Technical Abstract: Hyperspectral microscope imaging is presented as a rapid and efficient tool to classify foodborne bacteria species. The spectral data were obtained from five different species of Staphylococcus spp. with a hyperspectral microscope imaging system that provided a maximum of 89 contiguous spectral images using acousto-optic tunable filters (AOTF) platform. Spectral fingerprints were extracted from bacterial samples at a cell level using the region of interest on the cells. With a Mahalanobis distance algorithm, the outlier beyond 99% confidence of data was removed and then five species of staphylococcus (aureus, hyicus, sciuri, simulans, haemolyticus) were classified with partial least square discriminant analysis (PLS-DA) and support vector machine (SVM) discriminant methods for both linear and non-linear representative classification methods. With a PLS-DA method, five species were classified with 93.5% accuracy and 0.91 kappa coefficient; whereas, the classification accuracy was improved up to 98.2% with 0.97 kappa coefficient when the SVM algorithm was applied for same data. To develop classification model for near real-time application, only six wavelengths including 606, 462, 594, 590, 546, and 550 nm that are highly correlated with the classification accuracy were examined. In this case, however, the SVM classification accuracy decreased to 74.3% with 0.68 kappa coefficient.