Deep-learning-assisted label-free single-cell identification and sorting using femtosecond laser fabricated microfluidic surface-enhanced Raman scattering (SERS) chips with plasmonic ring-shaped nanostructure arrays

Accurate and efficient isolation and recovery of interest cells allow detailed analyses of cellular function, growth and culture, and gene-expression profiling of cells within heterogeneous populations. We developed a novel microfluidic surface-enhanced Raman scattering (SERS) chip capable of performing deep-learning assisted, label-free single-cell identification and sorting. Because both high sensitivity and spatially uniform resolution of the SERS substrate are critical for label-free cell sorting, laser near-field reduction of Au ions, which enables precision fabrication beyond optical diffraction limits, was employed to fabricate periodic plasmonic nanostructure arrays. The fabricated SERS substrates exhibited high analytical sensitivity (analytical enhancement factor: ~106) and uniform spatial enhancement (relative standard deviation: ~2%) with a spatial resolution of ~185 nm. We demonstrate for the first time the importance of high spatial resolution in achieving label-free cell identification with high accuracy (~96.2%) and AI classification sensitivity (~97.1% recall) using deep-learning-based analysis of Raman mappings. The utility of the fabricated system in cell sorting was confirmed by integrating it into a functionalized microfluidic device for cancer-cell sorting. The results could be used to guide future research on the design and development of non-invasive, efficient cell-sorting techniques with potential applications in clinical diagnostics and precision medicine.