The evaluation of foodborne pathogenic bacteria using near infrared (NIR) hyperspectral imaging and multivariate image analysis

[摘要] ENGLISH ABSTRACT: Near-infrared (NIR) hyperspectral imaging (HSI) and multivariate image analysis (MIA) was investigated for its potential as a rapid analytical method for the identification of foodborne pathogenic bacteria and distinguishing between the various genera and species used.NIR hyperspectral images of Bacillus cereus, Escherichia coli, Salmonella enteritidis, Staphylococcus aureus and a non-pathogenic bacterium, Staphylococcus epidermidis were acquired using a sisuChema SWIR (short wave infrared) hyperspectral pushbroom imaging system with a spectral range of 920-2514 nm. Hyperspectral images of streaked out (on Luria Bertani agar) bacteria were acquired after 20, 40 and 60 h growth (37 °C). Principal component analysis (PCA) was applied to mean-centered data, and was used to remove background and bad pixels from images. To investigate the possibility of distinguishing between bacteria, standard normal variate (SNV) correction and the Savitzky-Golay technique (2nd derivative, 3rd order polynomial; 25 point smoothing) was applied to data of growth plates imaged after 20 h. PCA score plots, score images and loading line plots were then evaluated. Bacteria were divided into 3 groups which were merged into mosaics. One group contained bacteria which appeared similar in colour (white) on the growth media (B. cereus, E. coli and S. enteritidis), another contained 3 Gram positive bacteria (B. cereus, S. aureus and S. epidermidis) and the third contained 2 species of the same genera (S. aureus and S. epidermidis). On the cleaned images, PCA score plots illustrated distinct chemical differences between colonies which appeared similar in colour on growth media. It was possible to distinguish B. cereus from E. coli and S. enteritidis along PC1 (58.1 % SS) and between E. coli and S. enteritidis in the direction of PC2 (7.75 % SS). S. epidermidis was separated from B. cereus and S. aureus along PC1 (37.5% SS) and was attributed to variation in amino acid and carbohydrate content. Two clusters were evident in the PC1 vs. PC2 PCA score plot for the images of S. aureus and S. epidermidis, thus permitting distinction between species. Differentiation between genera, Gram positive and Gram negative bacteria and pathogenic and non-pathogenic species was possible using NIR hyperspectral imaging.Partial least squares discriminant analysis (PLS-DA) was used as the supervised classification method to differentiate and predict the types of bacteria present. For models built and tested using the 20 h growth plates, the best predictions were made of B. cereus and the two Staphylococcus species, where results ranged from 82.0-99.96% correctly predicted pixels. However the poorest predictions were made of E. coli and S. enteritidis where results ranged from 2.34-53.9. To improve on these results, the effect of colony age has on prediction accuracies were investigated, while keeping rapidity in mind. PLS-DA models were built on standard normal variate (SNV) treated data for 20, 40 and 60 h growth plates and tested on a second set of plates. Predictions were improved – the highest overall prediction accuracies, where test plates required the least amount of growth time, was found with models built after 60 h growth and tested using 20 h growth plates. Predictions for bacteria differentiation within these models ranged from 83.1 to 98.8 % correctly predicted pixels.

[发布机构] Stellenbosch University