Based on the complementary and enhanced fusion of 3D point clouds and 2D RGB images, this paper designs an end-to-end learning framework-Point Cloud Enhanced Depth Pixel Fusion Network (PEPF-Net), aimed at enabling robots to achieve accurate 3D perception of unstructured environments. In the process, we address four key problems in 3D perception tasks: enhancing RGB representation using the reflection intensity and depth information of point clouds to generate Depth-RGB Pixel (D-Pixel); proposing Point-by-Point Vector Attention (PVA-Net) to model the vector relationships of point clouds,  to obtain deep-level point cloud features, and to achieve direct and effective fusion of heterogeneous data; designing a Layered-Transformer (L-TsfmNet) feature extractor to hierarchically extract D-Pixel features; proposing Variable Window Self-attention (VS-a) to focus on the relationships between local "window tokens" and avoid the complexity of global computation. Extensive experiments on the KITTI dataset demonstrate that PEPF-Net outperforms the currently common advanced environmental 3D perception algorithms.
The rapid growth of AI-driven applications in hybrid cloud-edge environments poses substantial challenges to ensuring low latency, high throughput, and effective resource utilization. Conventional deployment models, which are typically fixed or policy-driven, are not sufficiently flexible to respond dynamically to changing workloads and heterogeneous hardware environments. In this work, we introduce and analyze a resource-conscious deep learning-based scheduling system for managing the deployment of AI models on distributed cloud edges. The framework improves inference performance by leveraging real-time system telemetry and model features generated by benchmarks, while maintaining quality of service (QoS) compliance. The proposed system uses a fully connected neural network trained on structured features derived from the MLPerf Inference Benchmark, including compute complexity, memory footprint, and input dimensions. It is guided by real-time data from a hybrid infrastructure (NVIDIA A100/V100 GPUs and Jetson Xavier edge devices) to inform scheduling. Four MLPerf inference workloads - ResNet 50, BERT, SSD ResNet34, and DLRM - were tested and contrasted across various batch sizes and latency thresholds. Generalization experiments with unseen models such as GPT 2 and YOLOv5 yielded > 90% success rates in deployment, with the latency reduction and throughput gain results as presented above. Results of the generalization experiments with unseen models, including GPT 2 and YOLOv5, demonstrated deployment success rates > 90% for the various profiling conditions evaluated, with the latency reduction and throughput improvements as shown above. The results show that learning-based orchestration can be used to deliver space- and resource-aware orchestration solutions that are adaptive for low-latency deployment of AI services in hybrid cloud edge systems, but the effectiveness of the solution will depend on the representativeness of the profiling data and similarity of training and deployment environments.
Dynamic LiDAR point cloud compression with range images aims to reduce storage and transmission costs while preserving both spatial accuracy and temporal consistency across frames. Vision Transformers (ViTs) are commonly used for cross-frame dependency modeling. However, they suffer from feature misalignment under cross-frame displacement due to fixed patch partitioning, and their global attention across all patches is costly yet ineffective for local motions. High-precision sequences also face precision loss when 16-bit range data are quantized in a single channel. To address these limitations, we propose a Slide Deformable Transformer framework for high-precision dynamic LiDAR point cloud compression, termed SDT-PCC. At its core, the proposed SDT layer restricts attention to local sliding windows, capturing fine-grained correspondences across consecutive frames. It integrates deformable convolution into cross-frame attention to adaptively sample motion-offset locations, thereby enhancing temporal alignment and motion modeling. We also propose a Radix-Decomposition Multi-Channel Quantizer (RDMCQ), which decomposes range values into multiple channels and progressively refines precision across radix levels. Consequently, these designs can produce more temporally-coherent, accurate and stable reconstructions. Experiments on the SemanticKITTI dataset show that SDT-PCC achieves high efficiency in dynamic point cloud compression. The code is available on https://github.com/SYSU-SAIL/SDT-PCC.
Rapid activation of the cardiac catheterization laboratory (CCL) for ST-segment elevation myocardial infarction (STEMI) is essential to minimize time to reperfusion. However, system-wide efforts to reduce treatment delays have been accompanied by increased false activations, defined as activations that do not result in emergent coronary intervention. False activations contribute to unnecessary team mobilization (UTM), staff fatigue, workflow disruption, and inefficient resource use. This study aimed to evaluate whether the implementation of a cloud-based care coordination platform (Stenoa) was associated with reductions in false activations and UTMs at a high-volume tertiary cardiac center. In September 2021, the McGill University Health Centre implemented Stenoa, a mobile, cloud-based STEMI coordination platform enabling systematic case validation using electrocardiographic and clinical data. A retrospective cohort study was conducted, including all CCL activations between September 2020 and December 2022. Activations were grouped as preimplementation (group 0: September 2020 to September 2021) and postimplementation (group 1: September 2021 to December 2022) periods. A false activation was defined as a CCL activation followed by case cancellation before any procedure was performed. The primary outcome was the rate of UTM. In total, 632 activations were analyzed (group 0: n=288; group 1: n=344). UTM decreased from 8.7% (23/265) to 4.4% (14/316) following platform implementation (P=.04). False activation frequency decreased from 10.2% (27/265) to 6.9% (22/316), although this difference did not reach statistical significance (P=.16). Among false activations, the proportion resulting in UTM declined from 85% to 63% (P=.08). The implementation of a cloud-based STEMI coordination platform was associated with a significant reduction in unnecessary catheterization laboratory team mobilization. Structured digital communication may improve workflow efficiency and resource use in STEMI systems of care. Further multicenter evaluation is warranted.
To address the core bottlenecks in the evolutionary performance evaluation of automobile door stamping processes, namely insufficient adaptation to dynamic uncertain information and the lack of systematic methodological support in existing methods, this study establishes a complete evaluation methodological framework covering the indicator system, weight algorithm, and evaluation model. First, based on the PDCA cycle logic, a multi-stage indicator system integrating static benchmarks and dynamic evolution dimensions is designed, covering the entire process of planning, execution, inspection, and optimization. Second, an entropy weight-information cloud coupled weighting algorithm is proposed. The entropy weight method is used to extract objective data features, and the fuzzy-random modeling capability of the information cloud model is combined to achieve robust weight allocation of evolutionary indicators. Finally, to solve the problem that the traditional TOPSIS method cannot distinguish the advantages and disadvantages of schemes near the ideal solution, JS divergence is introduced to improve the traditional TOPSIS algorithm, and the distance between the process scheme and the positive/negative ideal solutions is quantified to realize accurate performance ranking. Through case validation on automobile door stamping processes, five typical door stamping process schemes are evaluated. The results show that the comprehensive performance score ranges from 0.1765 to 0.8689, and the performance ranking is highly consistent with the actual production logic, verifying the effectiveness and industrial adaptability of the proposed methodology. This study provides a systematic quantitative tool for the evolutionary performance evaluation of processes in complex manufacturing scenarios, and has important theoretical reference and engineering application value.
Stem cell research offers unique opportunities for authentic scientific engagement, yet infrastructure requirements have confined participation to elite institutions, perpetuating workforce disparities. We developed an integrated framework combining stem cell engineering, cloud-connected microscopy, and psychometric assessment. The framework integrates a doxycycline-inducible NGN2 mouse embryonic stem cell (mESC) system, low-cost cloud microscopy, and the Stem Cell Research Identity Scale (SCRIS). Implementation across a high school and a community college demonstrated significant increases in scientific identity. Students using differentiating PSCs showed broader science identity gains than those using neuroblastoma cells, particularly in competence, research readiness, and recognition. High school students showed enhanced research competence gains compared to community college students despite equivalent intervention duration. Demographic analyses revealed enhanced effectiveness for Hispanic and first-generation college students. This framework provides a scalable model for broadening participation in biomedical research.
We introduce a framework for analysing topological tipping in time evolutionary point clouds by extending the recently proposed topological optimal transport (TpOT) distance. While TpOT unifies geometrical, homological and higher-order relations into one metric, its global scalar distance can obscure transient, localized structural reorganizations during dynamic phase transitions. To overcome this limitation, we present a hierarchical dynamic evaluation framework driven by a novel topological and hypergraph reconstruction strategy. Instead of directly interpolating abstract network parameters, our method interpolates the underlying spatial geometry and rigorously re-computes the valid topological structures, ensuring physical fidelity. Along this geodesic, we introduce a set of multi-scale indicators: macroscopic metrics (topological distortion and persistence entropy) to capture global shifts, and a novel mesoscopic dual-perspective hypergraph entropy (node-perspective and edge-perspective) to detect highly sensitive, asynchronous local rewirings. We further propagate the cycle-level entropy change onto individual vertices to form a point-level topological field. Extensive evaluations of physical dynamical systems (Rayleigh-van der Pol limit cycles, double-well cluster fusion), high-dimensional biological aggregation (D'Orsogna model) and longitudinal stroke fMRI data demonstrate the utility of combining transport-based alignment with multi-scale entropy diagnostics for dynamic topological analysis. This article is part of the theme issue 'Critical transitions and intelligent control in complex systems'.
This work presents an open-source device for acquiring, correcting, and reconstructing the spectral power distribution (SPD) of LED sources used in controlled environmental agriculture. Unlike direct measurement spectrometers, the system employs a low-cost multispectral sensor (AS7265x, 18 channels, 410-940 nm) to acquire sparse band-integrated data, which are subsequently processed through a two-stage machine learning pipeline to infer a dense SPD representation. The sensor is integrated into an embedded platform that performs spectral acquisition, processing, wireless transmission, and remote visualization. Comparison with a reference spectrometer revealed non-linearities and some minor limits to the agreement between sensor data and ground-truth spectra. To address this, a correction stage based on a multilayer perceptron (MLP) implemented with TensorFlow Lite Micro was developed, reducing the RMSE from 0.183 to 0.035 and improving the reliability of the data. Complementary environmental monitoring was included using a BME688 sensor to record temperature, humidity, and gas concentration, serving as a reference to detect and correlate anomalies in SPD measurements under extreme environmental conditions. All data were transmitted to a back-end server for processing. Spectral reconstruction was performed in the cloud using a one-dimensional convolutional neural network (1D-CNN) trained on horticultural LED spectra and physically inspired synthetic spectra representative of CEA. The model achieved an RMSE of 0.0135, confirming high precision within the target application domain and demonstrating a scalable and cost-effective solution for spectral monitoring in controlled agricultural environments.
Cancer of Unknown Primary (CUP) remains one of the deadliest diagnostic challenges in oncology, accounting for 3-5% of all cancer diagnoses in the United States. Tumor origin is essential for guiding treatment; however, histological methods often fail in metastatic cases due to an inconclusive origin, leading to toxic and ineffective empirical therapies. Somatic mutations offer a promising alternative for accurate diagnosis via genomic alterations. We developed TumorOriginPredictor, a machine learning platform trained on 10,945 MSK-IMPACT mutation profiles. Five models generate the top three predicted tumor origins with rank-ordered probabilities, ensuring interpretability and clinical trust. Evaluated as a single platform, TumorOriginPredictor achieved over 72% top-3 accuracy across 12 cancer types and exceeded 80% for prevalent cancers. Clinically validated with 770 real-world profiles and deployed on Azure cloud, it provides real-time predictions, promoting individualized treatment by eliminating empirical therapies, reducing costs and time, and improving survival through a scalable, trustworthy AI platform.
The need for sophisticated and dependable prognostic systems for Induction Motors (IM) working under dynamic load and speed circumstances has increased due to the quick expansion of Electric Vehicles (EVs). Real-time defect prediction is limited by traditional multi-sensor data fusion techniques like concatenation, weighted averaging, or rule-based fusion, which are unable to determine which sensor is more informative at any given time. Additionally, early deterioration signs are missed by traditional models because they are unable to capture long-range temporal correlations across heterogeneous data. Additionally, these models have vanishing-gradient problems, which cause little but significant changes to be overlooked. An innovative prognostic learning system for integrated defect detection and remaining useful life (RUL) prediction in electric vehicle induction motors is presented in this work. It is based on Temporal Attention Fusion Long Short-Term Memory (TAF-LSTM) architecture. The suggested method successfully combines multi-domain feature extraction, long-sequence modeling, and temporal attention mechanisms to capture contextual fluctuations, gradual degradation patterns, and transient fault signals across several sensor channels. 99.9% fault diagnosis accuracy is demonstrated by experimental validation utilizing multi-sensor operating data. Additionally, for RUL prediction, the model obtains a Mean Squared Error (MSE) of 0.25 and a Root Mean Squared Error (RMSE) of 0.50, demonstrating extremely precise prognostic performance. These findings demonstrate that the suggested TAF-LSTM architecture provides a real-time, scalable, and dependable motor health monitoring solution with substantial promise for next-generation predictive maintenance in electric vehicles.
The growing number of smart devices supporting bandwidth-intensive and latency-sensitive applications, such as real-time video analytics, smart sensing, Extended Reality (XR), etc., necessitates reliable indoor wireless connectivity. In such environments, accurate Radio Environment Maps (REMs) enable adaptive wireless network planning and optimization of Access Point (AP) placement. However, generating realistic REMs remains difficult due to the variability of indoor environments and the limitations of existing modelling approaches, which often rely on simplified layouts or synthetic data. These challenges are further amplified by the adoption of next-generation Wi-Fi standards, operating at higher frequencies with limited range and wall penetration. To support progress in this area, we collected a dataset that combines high-resolution 3D LiDAR scans with Wi-Fi RSSI measurements across 20 setups in a multi-room indoor environment. It includes two measurement scenarios, one with and one without human presence, enabling development and validation of REM estimation models that incorporate physical geometry and environmental dynamics. The described dataset supports research in data-driven wireless modelling and the development of high-capacity indoor communication networks.
In medically complex patients with chronic pain, new or worsening symptoms should prompt careful reassessment. Avascular necrosis, particularly in the setting of prolonged steroid use, may be overlooked due to diagnostic overshadowing. Timely recognition requires multidisciplinary collaboration and vigilance to distinguish acute pathology from baseline chronic conditions.
This study suggests a novel extraction pipeline based on terrestrial laser scanning across multiple growth stages to address the current deficiency of three-dimensional (3D) phenotypic traits for wheat populations derived from 3D point clouds. This study presents 3D Wheat Point-seg Net (3D WP-seg Net), a novel 3D point cloud segmentation network that incorporates an SA-CrossAttention module to address the difficulties presented by complex structures, background noise, non-uniform point distributions, and scale variations in plot-level wheat point cloud data. Plot height, canopy area, and volume are examples of common phenotypic parameters that are successfully extracted using this technique. Additionally, two new phenotypic parameters: plot extension distance and lodging angle are suggested by fusing the centroid and slice-skeletonization algorithms. A software platform called 3D Trait Analysis was created to facilitate multi-sensor 3D data processing and trait extraction. A genome-wide association study (GWAS) was then conducted using the extracted population-level traits to find potential genes linked to these new phenotypes. While the segmentation accuracies of 3D WP-seg Net achieved 93.1%, 88.3%, and 92.5% under various sensor systems, the results showed a strong correlation between the predicted and measured plot heights (R2 = 0.954). Furthermore, four candidate genes linked to extension distance were found on chromosomes 1A, 2A, and 4A, and five putative genes controlling plot lodging angle were found on chromosomes 2D, 3A, and 7A. The multi-stage 3D phenotyping and analysis framework for wheat populations established by this study improves the accuracy of point cloud segmentation and trait quantification while offering a new and efficient method for the genetic analysis of important population-level traits.
To evaluate the accuracy of a high-precision navigation system that projects a three-dimensional (3D) kidney model onto surgical images by limiting its application to moments of minimal organ deformation during partial nephrectomy (PN). We analyzed 29 patients who underwent PN at Kyoto University Hospital and Kobe City Medical Center General Hospital in Japan. 3D models of the kidney and tumor were generated using DICOM data, whereas 3D point clouds of the surgical field were obtained using stereo camera recordings. Noise reduction processing was applied to the camera-derived point clouds. Registration between the computed tomography-derived models and camera-derived point clouds was performed using the closest iterative point, and the accuracy was assessed using the root mean squared error. We evaluated the effects of the point-cloud surface area and camera-to-target distance on the registration accuracy. Without noise reduction, the median registration error was 2.33 mm, whereas noise reduction improved the accuracy by 1.83 mm. The accuracy was significantly higher when the camera-to-target distance was shorter, with and without noise reduction. The surface area was inversely correlated with the accuracy without noise reduction, but no significant correlation was observed with noise reduction. Focusing on moments with minimal organ deformation, we demonstrated that high-precision surgical navigation is achievable in PN using actual surgical recordings. This may contribute to improved tumor localization and the preservation of renal function.
The FORUM (Far-infrared Outgoing Radiation Understanding and Monitoring) mission will provide, for the first time, systematic far-infrared spectral measurements of Earth's outgoing radiation, enabling improved understanding of atmospheric processes and the radiation budget. Retrieving atmospheric states from these observations constitutes a high-dimensional, ill-posed inverse problem, particularly under cloudy-sky conditions where multiple-scattering effects are present. In this work, we develop a data-driven, physics-aware inversion framework for FORUM all-sky retrievals based on latent twins: coupled autoencoders for atmospheric states and spectra, combined with bidirectional latent-space mappings. A lightweight model-consistency correction ensures physically plausible cloud variable reconstructions. The resulting framework demonstrates potential for retrievals of atmospheric, cloud and surface variables, providing information that can serve as a prior, initial guess, or surrogate for computationally expensive full-physics inversion methods. It also enables robust scene classification and near-instantaneous inference, making it suitable for operational near-real-time applications. We demonstrate its performance on synthetic FORUM-like data and discuss implications for future data assimilation and climate studies.
High-fidelity 3D reconstruction and precise phenotypic parameter extraction of banana plants are critical for crop growth monitoring and yield estimation in precision agriculture. However, traditional methods encounter significant bottlenecks: LiDAR systems are cost-prohibitive for widespread adoption, while traditional photogrammetry often fails to handle the complex canopy structures, severe occlusions, and weak texture features characteristic of banana leaves. To address these limitations, this article proposes a novel framework for 3D reconstruction and automatic phenotyping based on multi-view images captured by mobile phones. We introduce BN-NeRF, an enhanced Neural Radiance Field method built upon Instant-NGP. Specifically, we integrate three key technical improvements: (1) frame-level geometric calibration to correct camera pose drift caused by handheld motion; (2) sparse geometric anchoring to explicitly constrain depth and scale using sparse point clouds; and (3) thin-leaf prior regularization to suppress artifacts and improve the geometric accuracy of leaf surfaces. Building on this reconstruction, we establish a complete pipeline to recover explicit metric geometry from implicit radiance fields. By combining mesh topological analysis with geodesic algorithms, we achieve automated and precise extraction of key morphological parameters. Extensive experiments were conducted on a dataset of 90 banana plants in a real-world orchard. The results demonstrate that BN-NeRF achieves superior rendering quality (PSNR of 32.4 dB, SSIM of 0.951, and LPIPS of 0.152) while maintaining inference speeds comparable to Instant-NGP. Furthermore, the extracted phenotypic parameters showed strong agreement with manual ground truth across both leaf-level and structural traits. In addition to trait-specific regression performance, the evaluation also includes normalized completeness analysis, calibration-cube-based scale validation, and Bland-Altman agreement analysis, supporting the measurement reliability of BN-NeRF for field phenotyping. This study demonstrates that low-cost smartphone-based acquisition, combined with BN-NeRF, can support accurate field phenotyping of banana plants. In addition, an implemented mobile-cloud system was functionally validated through repeated end-to-end runs on an iPhone 13 client and a cloud workstation.
Distinguishing self from non-self is a major challenge for the immune system. Endogenous cytoplasmic double-stranded RNA (dsRNA) can mimic viral RNA and activate immune sensors like MDA5. ADAR1-mediated adenosine-to-inosine editing disrupts base-pairing to suppress immunogenicity of these endogenous structures. Global editing indices are widely used to probe this crucial ADAR1 function. However, they are dominated by nuclear pre‑mRNA edits with limited immune relevance. Here we present the cytoplasmic editing index (CEI) that quantifies editing specifically within inverted Alu repeats in 3' untranslated regions of mature cytoplasmic transcripts, which potentially form cytosolic dsRNA structures carrying higher immunological risk. Analyzing over 25,000 RNA-sequencing samples, we demonstrate CEI captures ADAR1p150 activity and outperforms the global editing index in terms of sensitivity and signal-to-noise, enabling sharper tissue-specific profiling, enhanced detection power of infection‑induced editing changes, and stronger association with cancer prognoses. An open-source, cloud-native pipeline delivers end‑to‑end, reproducible analysis at very low cost, supporting immediate, scalable adoption. CEI provides a refined metric for quantifying immune-relevant RNA editing, revealing previously obscured tissue- and disease-specific editing landscapes. The accompanying open-source, cloud-native pipeline enables broad adoption of high-quality editing analysis across research settings. This approach offers new opportunities for investigating ADAR1's role in immunity, infection, and cancer, with potential applications in biomarker development and therapeutic intervention strategies.
This article reviews the early history of our solar system from an astrobiological perspective and presents evidence from meteorites and astronomical observations. The purpose is to trace the formation of key molecules that participated in the building blocks of life. The Sun and its planetary system started from a section of a molecular cloud that collapsed into a protoplanetary disk. In the center of the protoplanetary disk, the protosun heated the surrounding material. The dust and gas inherited from the cloud remained pristine farther away from the protostar, while new compounds were created in the gas and on the icy mantles of the dust. The dust accreted into pebbles, pebbles formed planetesimals, and planetesimals collided and accreted pebbles to create planets. Meanwhile, the protosun became the Sun when its core reached the pressure and temperature required to transform hydrogen into helium. During this process, the Sun emitted high-energy radiation and particles that impacted the chemistry in the disk and the early evolution of the terrestrial planets.
Quantum or quantum-inspired Ising machines have recently shown promise in solving combinatorial optimization problems in a short time. Real-world and practical applications, such as time division multiple access (TDMA) scheduling for wireless multi-hop networks, financial trading, and emerging in-vehicle systems, require solving those problems sequentially where the size and characteristics change dynamically. However, using Ising machines for practical deployment involves challenges to shorten system-wide latency due to the transfer of large Ising model or the cloud access and to determine the parameters for each problem. Here we show a combinatorial optimization method using embedded Ising machines, which enables solving diverse problems at high speed without runtime parameter tuning. We customize the algorithm and circuit architecture of the simulated bifurcation-based Ising machine to compress the Ising model and accelerate computation and then build a machine learning model to estimate appropriate parameters using extensive training data. In TDMA scheduling for wireless multi-hop networks, our demonstration shows that the sophisticated system can adapt to changes in the problem and has a speed advantage over conventional methods.
Eucalyptus supports Ethiopia's economy and its zero-carbon strategy, yet its rapid expansion in the highlands of Ethiopia creates ecological concerns. For better management, accurate mapping is needed, but it is challenged by cloud contamination, spectral mimicry, and the requirement for high-resolution commercial imagery. Therefore, this study used an integrated multi-sensor satellite data, such as Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 Multi-Spectral Imagery (MSI), for reliable and cost-effective mapping of Eucalyptus trees in Meket district, Ethiopia. Sentinel-2 data is endowed with multi-spectral bands that are sensitive to chlorophyll and leaf water content, whereas Sentinel-1 SAR also functions in all weather conditions and can capture moisture and structural information of trees. Eighteen features from spectral bands, radar backscatter, and vegetation indices were fused using a feature-level fusion strategy and classified using Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Trees (CART). Results show that using freely available satellite data, RF achieved the highest performance, with an Overall Accuracy (OA) of 90% and a kappa coefficient of 0.81 in detecting Eucalyptus trees. SVM also achieved nearly the same performance with a 1% difference from RF. The study concludes that using publicly available Sentinel-1/2 fusion data with an appropriate classifier provides cost-effective, reliable, and accurate results for mapping of Eucalyptus, and supports Ethiopia's zero-carbon strategy. It also helps policymakers and planners by providing geospatial technology-based land use planning for monitoring and sustainable management of Eucalyptus trees.