A Rapid and Reliable Method of Discriminating between Listeria Species Based on Raman Spectroscopy

Raman spectroscopy is a rapid, non-invasive measurement mode that is readily able to discern bacteria from different genera (e.g. E. coli vs. Listeria spp.). At the species level, however, the task is more difficult. Herein we present a method of processing surface-enhanced Raman scattering (SERS) spectra from a low-cost, portable Raman system to discriminate between Listeria species. For six species adsorbed onto silver colloidal nanoparticles, SERS spectra were preprocessed (consisting of wavelet denoising and baseline removal), then features were extracted by selecting significant detail coefficients from a stationary wavelet transform decomposition. Classification was performed using an LDA classifier in association with "leave one out" cross-validation. Classification accuracies of 95.8%, for the six category problem, and 98.2%, for a binary classification problem (L. monocytogenes vs. all others), were achieved. The features found to be the most discriminating (using a rank sum approach) were the wavelet coefficients in the same spatial vicinity as the raw spectral peaks.