Hyperspectral Imaging for Rapid Detection of Common Infected Bacteria Based on Fluorescence Effect.

The rapid and accurate detection of bacterial infections in wounds is crucial for clinical diagnosis. Traditional methods, such as bacterial culture and polymerase chain reaction (PCR), are invasive and time-consuming. In this study, we propose a non-invasive detection method for common bacteria in wound infections, combining fluorescence hyperspectral imaging (FHSI) with deep learning algorithms. FHSI technology captures fluorescence data from culture plates for eight bacterial species, extracting spectral features within the 420-700 nm wavelength range. To manage the complex spatial and spectral data, we developed a Spatial-Spectral Multi-Scale Attention Network (SSMA-Net). Our method achieves an impressive 98.52% accuracy in bacterial classification under various growth conditions and 98.71% accuracy in species-level identification, with classification possible at bacterial concentrations as low as 10⁴ CFU/mL. These results underscore the effectiveness of FHSI and deep learning for rapid, non-invasive bacterial typing, offering substantial potential for clinical applications.