Deep Learning-Assisted Light-Sheet Superresolution Sensing of Six-Dimensional Nanoparticle Dynamics in Living Cells

Although single-particle tracking (SPT) offers critical insights into nanoscale transport within living systems, conventional approaches, relying mainly on sequential image analysis, fundamentally struggle to extract high-dimensional dynamics from noisy optical signals. Here, we developed a deep learning (DL)-assisted spatiotemporal sensing framework for the direct inference of multidimensional single-nanoparticle (NP) dynamics from dual-view light-sheet scattering data in living cells. By integrating light-sheet superresolution (LSSR) sensing with a convolutional neural network (DL-LSSR), this method directly maps raw optical point-spread function patterns to a multidimensional-state vector, simultaneously resolving the three-dimensional position, rotational orientation [azimuth (φ), elevation (θ)], and instantaneous velocity of single anisotropic NPs. The DL-LSSR enabled the automated identification and classification of dynamic transport states, revealing transition behaviors within complex intracellular trajectories. Thus, by bypassing iterative processing, this data-driven approach achieved approximately 99-fold faster analysis than conventional methods. As a biological demonstration, this sensing framework was used to investigate the cellular uptake of fucoidan-conjugated gold nanorods in living cells, revealing an uptake-associated preferred orientation of ∼67° during membrane entry under the experimental conditions examined and identifying distinct motion regimes consistent with endocytic processes. This method provides a generalizable analytical strategy for the direct extraction of high-dimensional dynamic information from optical signals, providing a new paradigm for data-driven SPT that operates beyond the diffraction limit of light.