Automated Plasmonic Resonance Scattering Imaging Analysis via Deep Learning

Plasmonic nanoparticles, which have excellent local surface plasmon resonance (LSPR) optical and chemical properties, have been widely used in biology, chemistry, and photonics. The single-particle light scattering dark-field microscopy (DFM) imaging technique based on a color-coded analytical method is a promising approach for high-throughput plasmonic nanoparticle scatterometry. Due to the interference of high noise levels, accurately extracting real scattering light of plasmonic nanoparticles in living cells is still a challenging task, which hinders its application for intracellular analysis. Herein, we propose an automatic and high-throughput LSPR scatterometry technique using a U-Net convolutional deep learning neural network. We use the deep neural networks to recognize the scattering light of nanoparticles from background interference signals in living cells, which have a dynamic and complicated environment, and construct a DFM image semantic analytical model based on the U-Net convolutional neural network. Compared with traditional methods, this method can achieve higher accuracy, stronger generalization ability, and robustness. As a proof of concept, the change of intracellular cytochrome c in MCF-7 cells under UV light-induced apoptosis was monitored through the fast and high-throughput analysis of the plasmonic nanoparticle scattering light, providing a new strategy for scatterometry study and imaging analysis in chemistry.

Automated summary

The study introduces a high-throughput LSPR scatterometry technique using deep learning neural networks, which accurately extracts the scattering light of plasmonic nanoparticles in living cells. Compared to traditional methods, this technique exhibits higher accuracy and generalization ability, providing a new strategy for intracellular analysis.