The development of fast and reliable sensing techniques to detect food-borne microorganisms is a permanent concern in food industry and health care. For this reason, Raman microspectroscopy was applied to rapidly detect pathogens in meat, which could be a promising supplement to currently established methods.
In this context, a spectral database of 19 species of the most important harmful and non-pathogenic bacteria associated with meat and poultry was established. To create a meat-like environment the microbial species were prepared on three different agar types.
The whole amount of Raman data was taken as a basis to build up a three level classification model by means of support vector machines. Subsequent to a first classifier that differentiates between Raman spectra of Gram-positive and Gram-negative bacteria, two decision knots regarding bacterial genus and species follow. The different steps of the classification model achieved accuracies in the range of 90.6%–99.5%. This database was then challenged with independently prepared test samples. By doing so, beef and poultry samples were spiked with different pathogens associated with food-borne diseases and then identified. The test samples were correctly assigned to their genus and for the most part down to the species-level i.e. a differentiation from closely-related non-pathogenic members was achieved.