Xuanwen Hua

The rapid convergence of optical innovation and machine intelligence is reshaping biological imaging by enabling platforms that jointly advance image formation and computational reconstruction for highspeed, high-resolution volumetric microscopy. However, broadly accessible three-dimensional imaging at high spatiotemporal resolution remains limited by the reliance of existing supervised methods on large modality-matched training datasets, the computational burden of conventional iterative reconstruction, and sensitivity to optical mismatch arising from small deviations in the spatial-angular point spread functions. Here, we introduce HYPER-Net, a physics-conditioned self-supervised framework for Fourier light-field microscopy that integrates scan-free volumetric acquisition with fast, robust three-dimensional reconstruction. HYPER-Net incorporates experiment-specific point-spread functions into the learning process in two complementary roles: as the forward operator that enforces measurement consistency and as a conditioning signal that adaptively modulates intermediate feature representations. This design reduces reliance on paired experimental ground-truth volumes, improves robustness to system variation, and enables generalizable reconstruction across diverse biological contexts. Using human colon organoids, embryonic Xenopus laevis hearts, hiPSC-derived cardiac spheroids, and freely moving Caenorhabditis elegans , we demonstrate high-fidelity volumetric imaging of tissue morphology, cardiac function, calcium-contraction coupling, and locomotion-associated neural and muscular dynamics. These results position HYPER-Net as a versatile framework for rapid volumetric imaging and quantitative analysis of dynamic biological systems across basic research and biomedical applications.

Optical scattering in biological tissues fundamentally limits fluorescence imaging by disrupting spatial and angular information, thereby restricting volumetric visualization. Although hardware-intensive and computational approaches have advanced scattering microscopy, practical three-dimensional imaging through tissue remains constrained by instrumental complexity and axial ambiguity. Here, we present volumetric scattering microscopy (VSM), a scan-free, optical–computational framework for three-dimensional fluorescence imaging in scattering biological media. VSM captures angularly resolved speckle-encoded fluorescence using an aperture-segmented Fourier light-field configuration and reconstructs volumetric structure through adaptive feature-based descattering and joint sub-pupil alignment. This hybrid strategy preserves angular information embedded in scattered light without wavefront measurement or mechanical scanning, while maintaining the simplicity of a standard epi-fluorescence architecture. We demonstrate high-fidelity volumetric reconstruction across phantoms, engineered cellular systems, ex vivo tissues with volumetric muscle loss, and intact Xenopus embryos, achieving preserved spatial resolution, enhanced optical sectioning, and quantitative accuracy under strong scattering conditions. VSM supports large-field, robust volumetric imaging in both layered and fully embedded scattering environments. By transforming scattered light into a structured encoding resource, VSM establishes a scalable pathway toward routine three-dimensional fluorescence imaging in complex biological systems.

The rapid convergence of advanced microscopy and deep learning is transforming cell biology by enabling imaging systems in which optical encoding and computational inference are jointly optimized for volumetric information capture and interpretation. However, broadly accessible three-dimensional imaging at high spatiotemporal resolution remains constrained by volumetric reconstruction throughput, susceptibility to artifacts, and the burden of collecting modality-matched training data. Here, we introduce PAVR, a physics-aware light-field imaging platform that integrates single-shot volumetric acquisition with fast, end-to-end volumetric reconstruction. PAVR is trained entirely using in silico system responses, avoiding reliance on external high-resolution ground-truth modalities and enabling sample-independent reconstruction across diverse biological contexts. Using fixed and live mammalian cells, we demonstrate multicolor volumetric imaging of subcellular organelles, three-dimensional tracking of autofluorescent particles, and high-speed visualization of organelle remodeling and interactions. We further extend PAVR to quantify coupled morphological and functional dynamics in beating human induced pluripotent stem cell–derived cardiomyocytes under pharmacological perturbation. Together, PAVR establishes a scalable hardware–software platform for high-throughput volumetric imaging and quantitative analysis of dynamic cellular systems in both basic and translational settings.

This review distinguishes between scanning and nonscanning methods for 3D imaging flow cytometry (IFC) and evaluates their optical principles, trade-offs, and biological applications to outline future miniaturized, AI-enabled, autonomous platforms.

Super-resolution microscopy, particularly localization-based methods, necessitates careful balancing of optical complexity, computational demands, and user accessibility. Conventional strategies typically adopt either deterministic or learning-based approaches, overlooking opportunities to leverage their synergistic strengths. In this work, we introduce volumetric localization microscopy (VLM) with deep learning, a super-resolution methodology that integrates instrumental and algorithmic advancements for high-fidelity 3D single-molecule imaging. VLM employs a wavefront-optimized light-field configuration to capture single-molecule data, while a cascaded neural network reconstructs 3D volumes and extracts molecular coordinates at a 10 nm lateral and 25 nm axial localization precision with effective imaging depth over 4 µm. Unlike existing methods, VLM is trained exclusively with system-aware intrinsic point-spread functions, bypassing dependencies on external imaging modalities or sample-specific data training. We validate VLM across diverse biological specimens, demonstrating hardware simplicity, data efficiency, and minimal phototoxicity. We anticipate VLM will overcome current limitations in fluorescence microscopy, empowering broader advancements in biomedical research.

Three-dimensional imaging of biological specimens requires high spatiotemporal resolution, but the intensive computational demands of processing continuously recorded images hinder the real-time reconstruction of millisecond cellular events. Here, we introduce projection-estimation accelerated Richardson-Lucy (PEARL) Deconvolution, an accelerated 3D reconstruction technique that significantly reduces iteration requirements for image reconstruction based on Richardson-Lucy deconvolution. We validate PEARL's capabilities by imaging immuno-stained mitochondria, GFP-labeled peroxisomes, and human pluripotent stem cell-derived organoids, demonstrating enhanced reconstruction speed and accuracy across diverse biological applications.

Super-resolution microscopy, particularly localization-based methods, necessitates careful balancing of optical complexity, computational demands, and user accessibility. Conventional strategies typically adopt either deterministic or learning-based approaches, overlooking opportunities to leverage their synergistic strengths. In this work, we introduce deep volumetric localization microscopy (VLM), a super-resolution methodology that integrates instrumental and algorithmic advancements for high-fidelity 3D single-molecule imaging. VLM employs a wavefront-optimized light-field configuration to capture single-molecule data, while a cascaded neural network reconstructs 3D volumes and extracts molecular coordinates at a 10 nm and 25 nm localization precision in the lateral and axial dimensions, respectively, across an imaging depth exceeding 5 μm. Unlike existing methods, VLM is trained exclusively with system-aware intrinsic point-spread functions, bypassing dependencies on external imaging modalities or sample-specific data training. We validate VLM across diverse biological specimens, demonstrating hardware simplicity, data efficiency, and minimal phototoxicity. We anticipate VLM will overcome current limitations in fluorescence microscopy, empowering broader advancements in biomedical research.

Cardiovascular diseases remain a pressing public health issue, necessitating the development of advanced therapeutic strategies underpinned by precise cardiac observations. While fluorescence microscopy is an invaluable tool for probing biological processes, cardiovascular signals are often complicated by persistent autofluorescence, overlaying dynamic cardiovascular entities and nonspecific labeling from tissue microenvironments. Here we present multiscale recursive decomposition for the precise extraction of dynamic cardiovascular signals. Multiscale recursive decomposition constructs a comprehensive framework for cardiac microscopy that includes pixel-wise image enhancement, robust principal component analysis and recursive motion segmentation. This method has been validated in various cardiac systems, including in vitro studies with human induced pluripotent stem cell-derived cardiomyocytes and in vivo studies of cardiovascular morphology and function in Xenopus embryos. The approach advances light-field cardiac microscopy, facilitating simultaneous, multiparametric and volumetric analysis of cardiac activities with minimum photodamage. We anticipate that the methodology will advance cardiovascular studies across a broad spectrum of cardiac models.

Light-field microscopy (LFM) has revolutionized 3D imaging by enabling single-shot volumetric acquisitions of various biomedical targets. With the development of Fourier LFM, the image quality and computing efficiency have been further improved. However, the limitation of light-field approaches remains due to inevitable deconvolution artifacts and time-consuming reconstruction processes until the advent of deep learning. Deep learning methods offer significant advantages in the reconstruction process by learning complex mappings from the observed data to the underlying 3D structures, effectively handling noise and artifacts. The well-trained network can predict volumetric information from each light-field frame in nearly real-time. However, existing methods depend on large, high-quality experimental datasets from complex and expensive imaging modalities. Additionally, the risk of overfitting specific training data constrains the generalizability across different samples and imaging conditions. Here, we develop a novel approach to minimize the training resources and advance the light-field reconstruction with optics-aware deep-learning and in-silico training datasets. Specifically, we train the network with only simulated randomly distributed point-spread functions (PSFs). The training process only considers the optics of the system rather than the prior knowledge of targeted structures. Therefore, with one step of light-field PSF calibration ahead, the training dataset preparation and network training can be completed in silico. We demonstrate that this strategy enables the deep learning network to predict both point-like and more complex structures. We anticipate our approach has great potential for generalization to many other deep learning-powered microscopic techniques, thereby broadening its applicability in advanced biological research and diagnostics.

Transplantation of ex vivo engineered hematopoietic stem cells (HSCs) can lead to robust clinical responses but carries risks of adverse events from bone marrow mobilization, chemotherapy conditioning and other factors. HSCs have been modified in vivo using lipid nanoparticles (LNPs) decorated with targeting moieties, which increases manufacturing complexity. Here we screen 105 LNPs without targeting ligands for effective homing to the bone marrow in mouse. We report an LNP named LNP⁶⁷ that delivers mRNA to murine HSCs in vivo, primary human HSCs ex vivo and CD34⁺ cells in rhesus monkeys (Macaca mulatta) in vivo at doses of 0.25 and 0.4 mg kg⁻¹. Without mobilization and conditioning, LNP⁶⁷ can mediate delivery of mRNA to HSCs and their progenitor cells (HSPCs), as well as to the liver in rhesus monkeys, without serum cytokine activation. These data support the hypothesis that in vivo delivery to HSCs and HSPCs in nonhuman primates is feasible without targeting ligands.