Staphylococcus Aureus-Related antibiotic resistance detection using synergy of Surface-Enhanced Raman spectroscopy and deep learning

The pandemic of antibiotic resistance, particularly Staphylococcus aureus-related resistance, is increasing frighteningly. The recommendation of wrong antibiotic drugs for patients due to misdiagnosis is one of the main reasons for spreading the pandemic. Therefore, dependable and fast technology should be carried out to get accurate results. Surface-Enhanced Raman Spectroscopy (SERS) has proven its ability to detect bacteria. SERS combined with machine learning algorithms have been employed recently to classify bacterial species; however, some machine learning algorithms are not convenient for some bacterial SERS spectra. Therefore, the performances of the algorithms are not good enough to extend to clinical use. In this study, we have investigated U-Net and VGG-16 algorithms in the classification of methicillin-resistant and susceptible Staphylococcus aureus SERS dataset. Moreover, the bands that play a key role in the classification of the two groups were visualized by the Gradient-weighted Class Activation Mapping (Grad-CAM) method. In addition, we have used some traditional machine learning algorithms to compare their performances with deep learning models. The results show that SERS spectra of methicillin-resistant and susceptible S. aureus are successfully classified with accuracies of 99% and 98% by U-Net and VGG-16 models. These findings provide great potential for clinical use.