high-speed microscopy

Fluorescence microscopy videos are often compromised by noise, temporal variability, and signal oscillations, hindering accurate analysis of dynamic biological processes. This work proposes an interpretable, end-to-end computational framework that, for the first time, integrates multi-temporal image registration, feature alignment, and cross-domain interpretable visual algorithms to efficiently compress dynamic video sequences into a single high-quality image while preserving critical biological structures. Experiments on a complex dataset of cardiomyocyte monolayers demonstrate that the proposed method increases the average number of detected cells by 44% compared to existing approaches, significantly enhancing both image quality and downstream segmentation performance.

This work addresses the bottleneck in high-throughput video acquisition, where the readout and transmission bandwidth of a single chip struggles to keep pace with rapidly increasing pixel counts. The authors propose a compact, high-speed video camera architecture that, for the first time, integrates low-bit gradient imaging with a multi-scale convolutional neural network (CNN) for image reconstruction. This approach substantially reduces data volume while efficiently recovering high-resolution images. By leveraging the fast readout capability of gradient sensors and sub-micron pixel integration technology, the system effectively alleviates throughput and complexity constraints. Both simulations and real-world experiments demonstrate that the proposed method achieves excellent reconstruction quality while significantly enhancing system efficiency, enabling a compact single-chip design.

This work addresses the challenge of capturing three-dimensional information from high-speed dynamic scenes using conventional low-frame-rate cameras, which are inherently limited in temporal resolution. Existing high-speed imaging approaches often rely on specialized hardware modifications or are constrained to single-view setups, hindering multi-view high-speed volumetric reconstruction. To overcome these limitations, the authors propose a novel computational imaging paradigm that employs temporally coded colored illumination to encode high-speed motion into color channels. This enables synchronized standard low-speed, multi-view cameras—without any hardware modification—to capture the necessary data for reconstructing high-frame-rate 3D volumes. By integrating a dynamic Gaussian splatting algorithm to decode the spatio-temporal information, the method achieves multi-view high-speed 3D reconstruction solely with off-the-shelf cameras. Its efficacy and practicality are validated through both simulated and real-world multi-camera experiments.

Two-dimensional tissue sections struggle to reliably capture the spatial characteristics of local cellular interactions and rare cell populations within three-dimensional tissue architecture, while dense volumetric imaging remains prohibitively expensive. This study systematically evaluates, for the first time, the bias introduced by 2D sampling on local spatial statistics and proposes a geometry-aware sparse 3D reconstruction framework. By integrating phenotypic similarity with spatial proximity to associate cells across serial sections and incorporating cell type–specific shape priors, the method reconstructs high-fidelity single-cell 3D coordinates. Validated on both public imaging mass cytometry and in-house CODEX datasets, the approach significantly enhances the reliability of spatial analysis under limited imaging budgets, outperforming conventional 2D analyses. It enables structure-level 3D spatial resolution and provides quantitative guidance for experimental design regarding section spacing, coverage, and redundancy.

To address the bottleneck where data generation rates in high-throughput imaging systems (e.g., PRISM) vastly exceed real-time processing capabilities, this work proposes a scalable FPGA-based streaming preprocessing architecture. The method jointly optimizes DRAM access patterns, inter-frame subtraction, and mean filtering, leveraging AXI4 burst transfers and streaming buffers to achieve real-time denoising and compression within a single frame interval. Implemented via high-level synthesis (HLS), the hardware pipeline sustains PRISM-scale throughput (>10 Gbps). Experimental results demonstrate sub-frame end-to-end latency, 3.2× raw data compression, and substantial offloading of subsequent CPU/GPU analytics. The core contributions are a frame-rate-constrained, low-latency on-chip denoising architecture and an efficient DRAM scheduling mechanism optimized for streaming imaging workloads.

This work addresses the lack of low-cost, portable biochemical concentration detection devices in resource-constrained settings by proposing a smartphone camera–based imaging system. By integrating custom optical attachments with tailored image processing algorithms, the system establishes a quantitative mapping model between sample colorimetric features and target analyte concentrations, effectively transforming a standard smartphone into a portable analytical platform. The method achieves detection accuracy comparable to that of commercial instruments across diverse samples—including fluorescein, RNA Mango aptamer complexes, homogenized milk, and yeast—while substantially reducing costs and enhancing the feasibility of on-site diagnostics.

Current deep learning approaches exhibit limited generalization in microscopic image analysis, struggling to adapt across diverse biological specimens, imaging modalities, and analytical tasks, which often necessitates manual intervention. This work proposes the first ready-to-use, general-purpose framework that unifies segmentation, tracking, and counting tasks under a common matching formulation. By leveraging the powerful correspondence capabilities of pretrained latent diffusion models, the method achieves robust performance across varying experimental conditions without requiring extensive task-specific fine-tuning. The approach substantially lowers deployment barriers and demonstrates reliable accuracy across a wide range of imaging scenarios and biological targets, thereby advancing the broad adoption of automated bioimage analysis.

This study addresses the inherent trade-offs among resolution, field of view, and acquisition speed in biomedical imaging, along with the challenge of non-rigid deformation registration. We propose a generalizable deep learning framework based on optical flow to achieve real-time multimodal video stitching. Methodologically, an automated synthetic data generation pipeline is designed to facilitate rapid cross-modal adaptation. By integrating a fine-tuned optical flow model, synthetic deformation field training, and end-to-end pixel-level registration, the framework significantly enhances both training efficiency and generalization capability. Experimental results demonstrate that the proposed approach comprehensively outperforms existing baselines in accuracy, robustness, and speed across seven imaging modalities, successfully enabling real-time large-field-of-view visualization.

Drift in scanning microscopy induces spatial misalignment of signals, severely limiting the accuracy of quantitative measurements. This study extends orthogonal scan drift correction to multidimensional datasets, including spectroscopic images and diffraction patterns, proposing a general-purpose drift correction method that operates without prior structural models. By integrating affine and non-rigid deformation algorithms, the approach precisely recovers probe positions and enables accurate data resampling. Implemented as an open-source, GPU-accelerated software package, the proposed method achieves processing speedups of two to three orders of magnitude. Consequently, this work provides an efficient, automated, and routine drift correction solution for quantitative microscopy imaging.

This study addresses the challenging and tedious task of navigating micrometer-scale three-dimensional specimens under a microscope to locate regions of interest for imaging. It presents the first systematic comparison of three interaction paradigms—2D desktop, 3D desktop, and virtual reality (VR)—within a real-world micro-manipulation context, evaluating their performance through a user study in terms of task efficiency, usability, and completion rates. The results demonstrate that VR significantly outperforms both traditional 2D and 3D desktop interfaces, offering superior speed, enhanced user experience, and higher acceptance among participants. In contrast, the 3D desktop interface provides no substantial advantage over its 2D counterpart. These findings establish VR as a promising new interaction paradigm for three-dimensional navigation at microscopic scales and underscore its potential for scientific visualization and microscale manipulation tasks.