Single-Bacterium Screening of Oral Pathogens by Terahertz Near-Field Nanoscopy Derived Multiparameter

Accurate identification of oral bacteria is essential for assessing oral and systemic diseases. Conventional phenotyping methods and molecular biology techniques exhibit limitations, including low sensitivity and operational complexity. Here, we present a terahertz (THz) scattering scanning near-field optical microscopy (s-SNOM) strategy enabling single-bacterium discrimination among multiple oral species. Eight bacterial genera and five Streptococcus species were investigated. Simultaneous acquisition of terahertz near-field images and intensity-height curves allowed the extraction of attenuation parameters (K, A₁) and morphological parameters (height, area). Correlation analysis revealed that K and A₁ were significantly distinct between rod and coccus, showing inverse relationships with the height and area. Using these four parameters, machine learning models, including Random Forests (RF), Gaussian Process Classification (GPC), and Support Vector Machines (SVM), achieved accurate bacterial differentiation, successfully classifying 23 genus-level pairs (82% achieving 100% accuracy) and 6 Streptococcus species pairs (60% at 100% accuracy) within the current data set. THz imaging effectively visualized bacterial morphology, while machine learning validated the discriminative power of combined parameters. Furthermore, an interactive Shiny-based predictor integrating RF, GPC, and SVM models was developed for real-time demonstration of bacterial classification outputs. Among these, RF exhibited the highest proficiency in classifying Streptococcus aureus, GPC excelled with Limosilactobacillus reuteri and Porphyromonas gingivalis, while SVM was effective for P. gingivalis. This study demonstrates a label-free, high-precision strategy for rapid, single-cell-level identification of oral bacteria using THz near-field nanoscopy, offering a promising platform for microbial diagnostics and biosensing.