Separation-free bacterial identification in arbitrary media via deep neural network-based SERS analysis

Abstract

Universal and fast bacterial detection technology is imperative for food safety analyses and diagnosis of infectious diseases. Although surface-enhanced Raman spectroscopy (SERS) has recently emerged as a powerful solution for detecting diverse microorganisms, its widespread application has been hampered by strong signals from surrounding media that overwhelm target signals and require time-consuming and tedious bacterial separation steps. By using SERS analysis boosted with a newly proposed deep learning model named dual-branch wide-kernel network (DualWKNet), a markedly simpler, faster, and effective route to classify signals of two common bacteria E. coli and S. epidermidis and their resident media without any separation procedures is demonstrated. With outstanding classification accuracies up to 98%, the synergistic combination of SERS and deep learning serves as an effective platform for "separation-free" detection of bacteria in arbitrary media with short data acquisition times and small amounts of training data.

Keywords: Bacterial detection; Deep learning; E. coli; S. epidermidis; Surface-enhanced Raman spectroscopy.