Deep Learning-Assisted Rapid Bacterial Classification Based on Raman Spectroscopy of Bacteria Lysed by Acoustically Driven Fiber-Tip Vibration.

Rapid and accurate identification of bacterial pathogens is critical for effective clinical decision-making and combating antibiotic resistance. Surface-enhanced Raman spectroscopy (SERS) combined with machine learning (ML) offers a powerful method for rapid, label-free bacterial identification. Conventional methods rely on surface molecular structures for identification, yet the richer and unique spectral information from intracellular biomolecules is often masked by the bacterial envelope, limiting classification accuracy. Here, a novel bacterial classification method is demonstrated by introducing acoustofluidic lysis based on the vibrating fiber-tip, combined with Raman spectroscopy and deep learning. The fiber-tip oscillates in a torsional mode, generating a controlled single-vortex within a capillary to concentrate bacteria in high-shear regions, enhancing lysis efficiency. This process effectively exposes intracellular components such as nucleic acids, proteins, and lipids, significantly enhancing the expression of features in bacterial Raman spectra, improving both spectral resolution and information richness. A residual neural network (ResNet) model is further employed for automated classification, achieving 98.9% accuracy across seven bacterial samples, surpassing traditional classifiers like random forests. The clinical validation experiments highlight the method's potential for real-world applications, enabling direct, on-site detection of clinical samples and facilitating rapid diagnostics, thus offering a promising advancement in pathogen identification.