Rapid identification of live and dead Salmonella by surface-enhanced Raman spectroscopy combined with convolutional neural network

The presence of Salmonella in any ready-to-eat food is not acceptable. However, sub-lethal and dead pathogens may still exist in sterilized foods. Most detect methods can not differentiate live/dead bacteria. Therefore, it is necessary to differentiate live and dead Salmonella rapidly in sterilized foods, in order to prevent its spread and ensure food safety. In this study, surface-enhanced Raman spectroscopies (SERS) were used to detect Salmonella Typhimurium, Salmonella Enteritidis and Salmonella Paratyphoid and to distinguish between live and dead cells of all serotypes. Bacteria cells of three Salmonella serotypes were respectively prepared in 108 colony forming units /mL concentration and heated at 60 ? for different times up to 64 min when no live cell was left. In order to establish a fast identification method for live and dead Salmonella, the convolutional neural network (CNN) was proposed to recognize SERS spectra of live and dead Salmonella automatically. Fed with the SERS spectra, the average recognition accuracy of the stacking-CNN could reach 98.69 %, and the results showed that CNN is efficient for rapid identification of live and dead Salmonella, and is a less complicated methodology compared to traditional microbial methods.

Automated summary

Differentiating live and dead Salmonella in sterilized foods is necessary for ensuring food safety. In this study, surface-enhanced Raman spectroscopy and convolutional neural network were used to rapidly detect and identify live and dead Salmonella. The results showed that this method is efficient and relatively simple.