Rapid optical differentiation of bacterial species remains challenging in turbid biological media due to strong scattering and absorption. Here, we present a physics-informed, label-free biophotonic framework for bacterial identification based on refractive index (RI) spectral signatures. An opto-microfluidic PDMS chip integrating air mirrors, microlenses, and fiber-optic probes was developed to measure transmission spectra from eight clinically relevant ESKAPEE pathogens. To address the inverse problem of optical parameter retrieval, a hybrid modeling approach combining Monte Carlo photon transport, Mie scattering theory, and a genetic algorithm was implemented to reconstruct wavelength-dependent RI spectra. The reconstructed spectra reveal clear interspecies differences, with higher RI values for Escherichia coli and Klebsiella pneumoniae and lower values for Staphylococcus aureus. These physics-derived RI features were subsequently used for classification using a one-dimensional convolutional neural network, achieving 97.75% accuracy. This approach demonstrates the potential of RI spectral fingerprints for label-free microbial identification and on-chip biophotonic diagnostics.
Dinh et al. (Mon,) studied this question.