Optical Characterization of Gram-Positive and Gram-Negative Biofilms Using Swept-Source Optical Coherence Tomography and Deep Learning.

Biofilms are microbial communities enclosed in a matrix that increases antibiotic tolerance and underlies persistent device-associated infections, making label-free, non-destructive characterization important. Established differences in cell-wall structure between Gram-positive and Gram-negative bacteria are expected to alter light scattering, motivating the hypothesis that the two groups differ in optical attenuation. This proof-of-concept study tested that optical prediction using optical coherence tomography. The optical attenuation coefficient was extracted from cross-sectional images. Gram-positive biofilms showed higher attenuation than Gram-negative biofilms, with every cross-group comparison significant, while species within each group were statistically indistinguishable, indicating a Gram-level descriptor. Two deep-learning classifiers separated the groups with over 97% accuracy because the attenuation distributions overlap minimally; this demonstrates the feasibility of automation rather than a requirement for deep learning. The attenuation differences are consistent with the hypothesis presented in the study, suggesting that these structural differences may contribute to the observed optical contrast.