Machine Learning-Assisted Terahertz Metagrating Biosensor for Label-Free Bacterial Identification Based on Spectral...

Rapid and accurate identification of pathogenic bacteria is pivotal for enabling early clinical diagnosis and precision antimicrobial therapy. In this study, we developed a label-free bacterial detection platform that integrates terahertz (THz) metagrating with machine learning algorithms to identify specific pathogens through spectral fingerprint analysis. This approach uses all-dielectric THz metagrating fabricated via 3D printing to amplify wave-analyte interactions, generating distinct resonance-enhanced spectral signatures. Supervised machine learning algorithms were employed to discern subtle spectral variations, in which support vector machines achieved high performance in identifying eight pathogens with an accuracy of 97.50%. Furthermore, our approach exhibited exceptional performance on two newly recruited clinical cohorts, achieving overall accuracies of 99.29% and 96.75%, with AUC values exceeding 0.99 for all eight bacterial species. Notably, 100% accuracy was achieved for Escherichia coli, Staphylococcus aureus, Enterococcus faecalis, and Enterococcus faecium across all the data sets. Our platform achieves complete bacterial identification within 5 min postculture, offering the advantages of being simple, rapid, cost-effective (<$0.10 per test), and compatible with downstream analytical technologies, displaying robust potential as a universal tool for clinical microbial identification.