Machine Learning-Aided Polaronic Nanoantennas for Label-Free Bioimaging of Virus Monolayers

The rapid identification and diagnosis of viruses through saliva are highly desirable for molecular diagnostics but face significant challenges due to the low concentration of viruses and interference from other biomolecules, such as proteins and lipids. Here, we discovered that coupled modes of surface plasmon polaritons and surface phonon polaritons exhibit ultra-sensitivity to the infrared vibrations of nearby molecules. Remarkably, this sensitivity extends not only to the vibrational intensity but also to the vibrational characteristics of these molecules. This unique capability enables the coupled modes to overcome the aforementioned challenges, allowing for ultra-sensitive and accurate virus identification in complex biological environments. As a proof of concept, we conducted hyperspectral bioimaging of severe acute respiratory syndrome (SARS)-related coronaviruses in saliva. With machine learning algorithms, we achieved a detection limit at the molecular monolayer level, with an accuracy of 93% and a detection time of just 10 minutes. Notably, this is the first instance of label-free imaging of SARS virus molecular monolayers.