Scalable Plasmonic Metasurface-Enabled Physics-Guided Self-Supervised Cellular Imaging

Abstract

High-quality cellular imaging, especially in live cells, remains constrained by the trade-off among signal-to-noise ratio, phototoxicity, and instrumentation complexity. Here, we report a scalable plasmonic metasurface that generates a spatially ordered array of fluorescence-enhancing near-field hotspots and enables self-supervised denoised, cellular imaging with improved feature readability on a conventional wide-field microscope. The registered hotspot lattice serves as a physics-derived functional prior that identifies where fluorescence amplification is physically grounded and steers neural-network training accordingly, reducing reliance on paired ground truth, large external pretrained models, or extensive supervised datasets. We demonstrate two labeling-density-dependent operating regimes: dense labeling for cytoskeleton structural imaging and sparse labeling for multiplexed sensing of plasma-membrane-associated dynamics across the hotspot array. Our work unites scalable nanophotonic hardware and self-supervised computational imaging into a practical platform for structural bioimaging and on-chip live-cell biosensing under simple wide-field imaging conditions.