Depth map filtering method for shape-from-focus recovery based on hybrid network model.

We aim to address the technical limitations in 3D respiratory estimation and image reconstruction from ultra-sparse views, overcoming data-acquisition constraints that often hinder conventional deformable image registration and volumetric imaging in real-time clinical applications. &#xD;We propose a Latent Accelerated Diffusion framework for Deformation Estimation enabling Real-time volumetric imaging (LADDER) framework, which integrates: (1) a deformation network (VoxelMorph) that generates a patient specific baseline deformation vector field (DVF) from pre treatment imaging, and (2) a latent diffusion model (LDM)-based DIR model that estimates DVF scaling factors and residual corrects to generate intra-treatment real-time DVF and dynamic volumetric images. The LDM compresses the baseline DVF into a compact latent manifold, enabling fast, projection conditioned refinement guided by anatomical cues. A physics informed loss enforces anatomical regularity and consistency with measured projections. LADDER was trained on the Learn2Reg dataset and evaluated on 10 DIR Lab lung datasets . &#xD;Main Results: With a dual projection input, an end-inhale to end-exhale baseline DVF and a compression and down-sampling factor of 8, on 10 test cases, LADDER achieves a mean target registration error (TRE) of 0.87±0.33 mm, and high volumetric structure similarity (3D SSIM > 0.95) and low volumetric reconstruction error (3D NMSE < 0.006), while maintaining real-time inference (~0.11-0.12s). Further analysis shows the range of DVF span impacts deformation accuracy, and the dual-projection input improves deformation fidelity and reduced variability across breathing phases compared to the single-projection. &#xD;Significance: LADDER enables real time 3D motion estimation and volumetric reconstruction from ultra sparse X ray views by conditioning diffusion in a patient specific DVF latent space. Its submillimeter TRE and real time speed indicate strong potential for next generation motion management and image guidance, supporting on the fly anatomical modeling to enhance the safety and efficacy of lung SBRT.

Accurate three-dimensional flow-field measurement is vital for diagnostics in aerospace, combustion, and thermal processes. We present a multi-view phase measurement deflectometry (PMD) method for volumetric temperature-field reconstruction. By encoding absolute screen coordinates in fringe phases, PMD determines ray directions on a pixel-wise basis, ensuring intrinsic geometric self-consistency between the deflection measurement and the 3D reconstruction without relying on cross-correlation operations. A chain-based joint calibration strategy with anchor cameras accommodates surround-view configurations lacking a global common field of view. The tomographic inverse problem is solved within an alternating direction method of multipliers framework incorporating total-variation regularization and visual-hull support constraints. Numerical simulations and experiments on a thermal plume validate the proposed method, with thermocouple comparison showing a mean absolute relative difference of 1.87% and a mean absolute difference of 26.3 K, offering a new, to the best of our knowledge, pathway toward high-resolution volumetric flow diagnostics.

Accurate and reliable 3D scene reconstruction is a key component of intelligent surgery, enabling enhanced spatial understanding and data-driven analysis in minimally invasive surgery (MIS). However, existing clinical systems are often bulky and workflow-incompatible, while vision-based Structure-from-Motion methods struggle with sparse textures and specularities, leading to unstable pose estimation and high computational cost. To address these limitations, we present SurGSplat++, a progressive, pose-free Gaussian splatting framework for monocular surgical scene reconstruction that requires no auxiliary hardware or pre-computed camera poses. Experiments show that SurGSplat++ achieves improved geometric stability, reduced pose drift, and superior novel-view synthesis compared with existing approaches. By producing accurate and consistent 3D reconstructions, the proposed method provides a practical solution for post-operative analysis, pre-operative planning, and data-driven surgical modeling in clinical environments. Code will be released at https://surgsplus.github.io/.

Unsupervised deep learning enables blur-free resolution enhancement in two-photon microscopy.

X-ray imaging serves as a fundamental tool for non-destructive inspection. Although conventional radiography is well suited for two-dimensional imaging, it cannot provide volumetric structure. Computed tomography provides three-dimensional reconstruction but remains constrained by bulky instrumentation, high radiation exposure, and cost. Here we demonstrate a patch-type scintillator integrated with multi-stage neural network that segments and reconstructs three-dimensional volumes from sparse angular two-dimensional radiographs. The scintillator is fabricated by electrospraying cellulose nanocrystals onto a bulk cellulose matrix, followed by dip-coating of perovskite, yielding a composite with enhanced radioluminescence under X-ray excitation. This flexible film conforms to complex geometries, enabling distortion-free and multi-angle imaging. Neural networks are trained on synthetic datasets and validated on experimentally acquired avian tibiotarsus radiographs, accurately reconstructing volumetric bone structures. This approach serves as a proof-of-concept for low-dose, accessible artificial intelligence-enabled three-dimensional X-ray imaging, demonstrating the feasibility of recovering macroscopic three-dimensional morphology from as few as three sparse projections.

Current generative design schemes for product styling are primarily limited to two-dimensional space, which fails to adequately represent the 3D volumetric attributes and hinders the direct translation of emotional intent into manufacturable forms. To overcome this limitation, this study proposes the ENM (Emotion data–NST–MVSNet) framework, a unified pipeline that integrates deep learning classification, style transfer, and multi-view stereo reconstruction. First, ResNet18 is employed to construct a large-scale emotional dataset, quantifying the implicit mapping between user emotional requirements and visual styling features. Second, Neural Style Transfer (NST) is utilised to generate emotionally compliant multi-view images while preserving semantic consistency. Third, an enhanced MVSNet, incorporating deformable convolutions and a distance-aware loss re-weighting strategy, is developed to robustly reconstruct these stylised images into high-fidelity 3D models, enabling a seamless transition from 2D emotional cues to 3D geometric forms. Validated through an automotive styling case study, the experimental results demonstrate that the proposed method achieves high reconstruction quality and aesthetic satisfaction, with user evaluation scores exceeding 4.3 across key emotional dimensions. This research provides a high-efficiency pathway for intelligent product styling design and lays a solid foundation for automated, emotion-driven 3D product prototyping and future cross-dimensional generative design.

High-density planting is an effective strategy to increase maize yield but imposes greater demands on plant architectural adaptability. To elucidate the structural response mechanisms of maize under varying planting densities, we developed a high-throughput 3D phenotyping system tailored to complex field conditions. High-precision point clouds of field-sampled plants were obtained via multi-view 3D reconstruction. Using a deep learning network, stem and leaf organs were semantically segmented (95.6% accuracy), while leaves were individually separated via clustering (94.8% accuracy). From these data, 31 plant architectural traits and 14 ear-leaf traits were extracted, establishing a hierarchical trait characterization system. Results showed that increased planting density significantly influenced plant architecture reshaping and structural coordination, leading to more compact plant forms and ear height position centralization. Ear leaves exhibited heightened sensitivity to density variation, particularly in leaf area, vertical distribution, and leaf inclination angle, suggesting an early-response role. Principal component analysis and clustering further revealed patterns of structural differentiation and key traits driving these changes under density treatments. The integrated workflow-comprising data acquisition, modeling, segmentation, clustering, trait extraction, and analysis-offers a robust approach for structural phenotyping and intelligent breeding selection in maize and other tall crops. This pipeline provides valuable technical support and data resources for optimizing dense planting strategies and advancing digital agriculture.

Generation of synthetic labeled datasets for anomaly detection in heritage architecture

Artificial intelligence for real-time motion tracking in MRI-guided radiotherapy: A systematic review.

This joint feature issue of Optics Express and Applied Optics is organized in conjunction with the 2025 Optica conference on 3D Image Acquisition and Display: Technology, Perception and Applications, which was held from 18-21 August as part of the 2025 Imaging and Applied Optics Congress in Seattle, United States. This feature issue presents 34 articles that cover the topics and scope of the 2025 3D Image Acquisition and Display conference. This introduction provides a summary of the articles published in this feature issue.

Internal anatomical motion challenges precise radiation delivery during external beam radiotherapy. Estimating and compensating for anatomical motion are essential for improving planned dose delivery to target volumes while sparing organs-at-risk. This research achieves accurate motion prediction using only planar X-ray imaging from conventional linear accelerators, without surrogate signals or invasive fiducial markers. We propose Deep-Motion-Net: a patient-specific end-to-end graph neural network (GNN) enabling 3D volumetric organ reconstruction from single in-treatment kV planar X-ray images at arbitrary projection angles. A 2D convolutional neural network (CNN) encoder extracts image features, which four feature pooling networks fuse to a 3D template organ mesh. A ResNet-based graph attention network then deforms the feature-encoded mesh. Training uses synthetically generated organ motion instances and corresponding kV images, created by deforming a reference CT volume aligned with the template mesh, generating digitally reconstructed radiographs (DRRs) at required angles, and applying DRR-to-kV style transfer via conditional CycleGAN. Quantitative testing on synthetic respiratory motion scenarios and qualitative assessment on in-treatment images from four liver cancer patients demonstrated overall mean prediction errors of 0.16±0.13 mm, 0.18±0.19 mm, 0.22±0.34 mm, and 0.12±0.11 mm across datasets. Mean peak prediction errors were 1.39 mm, 1.99 mm, 3.29 mm, and 1.16 mm. This approach leverages accessible in-treatment imaging, avoiding expensive MRI systems or invasive markers. To the best of our knowledge, this is the first deep learning framework reconstructing volumetric 3D organ models from single-view images at arbitrary angles throughout an entire in-treatment scan series. Our approach achieves sub-millimetre accuracy when validated on synthetic motion instances and demonstrates clinical feasibility on real-treatment kV images, for which volumetric ground truth is inherently unavailable. The code is available at https://github.com/isurusuranga/DeepMotionNet .

Multi-view 3D reconstruction, namely, structure-from-motion followed by multi-view stereo, is a fundamental component of 3D computer vision. In general, multi-view 3D reconstruction suffers from an unknown scale ambiguity unless a reference object of known size is present in the scene. In this article, we show that multi-view images captured using a dual-pixel (DP) sensor can automatically resolve the scale ambiguity, without requiring a reference object or prior calibration. Specifically, the defocus blur observed in DP images provides sufficient information to determine the absolute scale when paired with depth maps (up to scale) recovered from multi-view 3D reconstruction. Based on this observation, we develop a simple yet effective linear method to estimate the absolute scale, followed by the intensity-based optimization stage that aligns the left and right DP images by shifting them back toward each other using cross-view blur kernels. Experiments demonstrate the effectiveness of the proposed approach across diverse scenes captured with different cameras and lenses. Code and data are available at https://github.com/lilika-makabe/dp-sfm-tpami.git.

In the context of Shaoxing's in-depth promotion of the overall intelligent governance 2.0 digital reform, the construction of the "more enjoyable people's livelihood" service brand, and the construction of a common prosperity demonstration city, traditional social services face practical bottlenecks such as inaccurate demand identification, difficulty in scene reproduction, difficult to quantify effects, and uneven distribution of resources. Virtual reality (VR) technology is highly consistent with the needs of high-quality development of social services by virtue of its advantages such as immersive interaction, three-dimensional scene reconstruction, and real-time data collection. Based on the local reality of Shaoxing, this paper takes the precise, visual, and quantifiable transformation of the "three modernizations" as the core path, builds a VR application scenario system and quantitative evaluation mechanism that adapts to pension, disability assistance, minor protection, social work training, and grassroots governance.

Treatment-resistant schizophrenia (TRS) affects approximately 20-30% of patients, and a substantial proportion develop clozapine-resistant schizophrenia (CRS). Deep brain stimulation (DBS) has emerged as a potential neurosurgical intervention targeting dysfunctional cortico-striato-limbic circuitry. However, technical heterogeneity and limited clinical data constrain interpretation of outcomes. This systematic review was conducted in accordance with PRISMA guidelines and prospectively registered in PROSPERO (CRD420251080715). Seven electronic databases were searched from inception through January 31, 2025. Clinical studies investigating DBS in TRS or CRS were included. Methodological quality was assessed using Joanna Briggs Institute (JBI) tools. Data extraction emphasized stereotactic targeting methods, hardware configurations, stimulation parameters, and clinical outcomes. The total number of study is 6. We excluded paper by Manssuer 2023. The total study population comprises 21 patients. Seven studies involving 21 patients met inclusion criteria. Targets included the nucleus accumbens (NAcc; n = 11), substantia nigra (SNr; n = 1), habenula (HB; n = 2), subgenual cingulate (SCG; n = 4), subgenual anterior cingulate cortex (sgACC; n = 3). Reported PANSS total score changes ranged widely (11%-85.7%), reflecting substantial inter-individual variability and methodological limitations. Surgical complications occurred in 3 of 21 patients (14.2%), including infection and hemorrhage. All cases utilized open-loop stimulation and conventional cylindrical leads. Current evidence suggests a preliminary therapeutic signal for DBS in highly selected CRS patients, particularly with NAcc targeting. However, conclusions remain limited by small sample sizes, technical heterogeneity, and absence of controlled trials. Future investigations should prioritize standardized stereotactic reporting, volumetric lead reconstruction, and long-term safety assessment within specialized neurosurgical research settings. PROSPERO CRD420251080715.

Early detection of breast cancer significantly improves patient outcomes through timely and accurate diagnosis. This study proposes a hybrid image reconstruction method combining the Maximum Likelihood Expectation Maximization (MLEM) algorithm with the Distance Driven Method (DDM) for stationary digital breast tomosynthesis, aimed at producing high-resolution three-dimensional (3D) breast images. The method was initially validated using simulated projection data of a digital breast phantom modeled with two spheres of varying radii and attenuation coefficients. Fifteen projection images were generated over a 15° angular range (from +7° to –7° with 1° increments) to replicate realistic tomosynthesis acquisition. The focus plane was set 45 mm above the detector with a pixel size of 0.14 mm. Reconstruction results demonstrated enhanced image quality, with spatial resolution quantitatively evaluated using the Line Spread Function (LSF) across the smaller sphere. Compared to the Ray Driven Method (RDM), the MLEM-DDM approach provided better contrast, sharper edges, and fewer artifacts. Subsequently, the method was applied to experimental data from a real breast phantom. Volumetric reconstructions revealed detailed tissue structures across multiple planes, with spatial resolution assessed by line profiling through two aligned calcifications in the focus plane. The MLEM-DDM method preserved the shape and sharpness of these calcifications more effectively than MLEM-RDM. Overall, the proposed MLEM-DDM framework enhances spatial resolution and visualization quality in stationary digital breast tomosynthesis, demonstrating strong potential for improved early breast cancer detection and diagnostic accuracy

3D scene reconstruction from a limited number of viewpoints

Proprioception and reflexive control of muscle tone depend on the activity of muscle spindles, specialized sensory receptors embedded deep within skeletal muscle that detect changes in muscle length. Their location and complex three-dimensional architecture have historically limited morphological analysis to techniques such as silver-impregnation, muscle teasing, or serial sectioning followed by volumetric reconstruction. Here, we describe a workflow for three-dimensional, in situ visualization of muscle spindles in the rabbit tenuissimus muscle, a preparation uniquely enriched in spindles and well suited for whole-mount imaging. The protocol combines fluorescent labeling of spindle sensory and motor innervation, including intrafusal γ neuromuscular junctions labeled with α-bungarotoxin, with immunolabeling and solvent-based optical clearing. Optically cleared tenuissimus muscles were compatible with both whole-mount confocal and light-sheet microscopy, enabling volumetric imaging of complete spindle structures and detailed visualization of Ia annulospiral endings at the spindle equator. This approach provides access to spindle morphology and connectivity at multiple spatial scales while avoiding physical sectioning and reconstruction. By enabling reproducible three-dimensional imaging of intact muscle spindles, this workflow offers a practical platform for studying spindle structure and plasticity in health and disease.

Photogrammetry has been used consistently in forensic settings to document crime scenes three-dimensionally. Traditionally, a large number of still photos are taken, and information from those photos is used to create a three-dimensional (3D) model. This research aimed to determine how accurately 3DF Zephyr, a photogrammetry software, could calculate the position of a body-worn camera through recorded video. In 3DF Zephyr, frames are extracted at regular intervals from the video to create a 3D model. Videos were taken with the Axon Body 3, Body 2, and Flex 2 body-worn cameras, five trials per camera, and five measurements per trial. A 3D reconstruction of a mock scene was created from each body-worn camera with 3DF Zephyr. A FARO laser scanner was used to obtain ground truth data. Coordinates of the camera positions were recorded and compared to ground truth coordinates. The Axon Body 3, Axon Body 2, and Axon Flex 2 had average errors of 7.6 cm, 10.1 cm, and 9.5 cm, respectively. The total average error for all cameras was 9.1 cm with a standard deviation (σ) of 4.8 cm. Thus, it is possible to estimate camera position within a 24 cm radius circle, accounting for 98.7% of all errors. Estimating camera position allows an officer's position to be pinpointed within that circle. This technique can be used to corroborate witness testimony in officer-shooting incidents. This study's results may be limited, as accuracy can decrease under excessively low-light or low-texture conditions.

3D MR image acquisition is inherently time intensive, rendering it susceptible to patient motion during scanning. This may introduce significant blurring and artifacts, potentially necessitating reacquisition. We propose a modular framework to retrospectively correct for intrascan motion in 3D brain MRI, without active motion tracking. Serving as the backbone of our approach is an existing distributed and incoherent sampling scheme (DISORDER), combined with a fast network trained for highly undersampled reconstruction. This enables approximate reconstructions of anatomy after every few seconds, using only a tiny fraction of k-space data (< 2%). While these reconstructions are only approximate, we postulate they are sufficient to estimate motion patterns at said temporal resolution. Groupwise registration, notable for its elimination of registration bias, is utilized for estimating rigid motion parameters, which are leveraged to reconstruct the measured data with reduced motion artifacts. The approach was evaluated on 94 retrospectively and 3 prospectively motion-corrupted invivo 3D T1-weighted brain MRI acquisitions. The estimated motion parameters matched the known retrospective motion with 0.06 mm and 0.13° accuracy, resulting in an improvement in reconstruction quality from to SSIM for the retrospective scans. The prospective scans improved from to SSIM after correction in the case of gradual motion and from to SSIM for extreme motion. In conclusion, the proposed approach, that is free of external tracking devices or navigators, successfully estimated and corrected 3D motion between small subportions of a scan. This resulted in vastly improved image quality, making volumetric MRI substantially more tolerant to motion.