Towards Point-of-Care Infection Diagnosis Tool: A Multi-Spectral Fluorescence Approach for Bacterial Differentiation

Infected wounds pose a global challenge to healthcare practitioners and systems. Chronic wound infection is particularly concerning due to its common polymicrobial nature. The identification of bacterial species is traditionally done by time-consuming protocols such as culture and molecular DNA-based approaches, but new techniques have been proposed to improve time management. The natural emission of light by bacteria, the phenomenon of bacterial autofluorescence, has been proposed as a fast and non-invasive alternative. However, fluorescence spectra of different bacterial species often overlap, making distinction difficult at the species level. Hence, more efforts are needed to improve the accuracy of this approach. While current literature largely focuses on pathogenic bacteria, it neglects the distinction of commensal species, which also fluoresce. In this work, we aim to distinguish between 5 clinically relevant bacterial species, both pathogenic and commensal of the skin flora, relying on a small number of wavelength pairs. Principal Component Analysis was applied to reduced dimensionality and then ensemble learning was performed to predict different bacterial populations in polymicrobial scenarios. Moreover, our approach eliminates the need for absolute fluorescence intensity measurements, which may be misleading in clinical practice.