Search Results (454)

Predictive process monitoring estimates the future behaviour of running process instances based on historical event logs, with typical tasks including next-activity prediction, remaining-time estimation, and risk assessment. Existing recurrent and Transformer-based models achieve strong accuracy on individual logs but transfer poorly across processes and underuse the rich graph structure of event data. This paper introduces ProcessGFM, a domain-specific graph pretraining prototype for predictive process monitoring on event graphs. ProcessGFM employs a hierarchical graph neural architecture that jointly encodes event-level, case-level, and resource-level structure and is pretrained in a self-supervised manner on multiple benchmark logs using masked activity reconstruction, temporal order consistency, and pseudo-labelled outcome prediction. A multi-task prediction head and an adversarial domain alignment module adapt the pretrained backbone to downstream tasks and stabilise cross-log generalisation. On the BPI 2012, 2017, and 2019 logs, ProcessGFM improves next-activity accuracy by 2.7 to 4.5 percentage points over the best graph baseline, reaching up to 89.6% accuracy and 87.1% macro-F1. For remaining-time prediction, it attains mean absolute errors between 0.84 and 2.11 days, reducing error by 11.7% to 18.2% relative to the strongest graph baseline. For case-level risk prediction, it achieves area-under-the-curve scores between 0.907 and 0.934 and raises precision at 10% recall by 6.7 to 8.1 percentage points. Cross-log transfer experiments show that ProcessGFM retains between about 90% and 96% of its in-domain next-activity accuracy when applied zero-shot to a different log. Attention-based analysis highlights critical subgraphs that can be projected back to Petri net fragments, providing interpretable links between structural patterns, resource handovers, and late cases. Full article

Objectives: Strong cervical musculature is recognized as a protective factor against sports-related concussions. Evidence suggests that jaw clenching may activate cervical muscles, potentially reducing head acceleration during impact. Methods: This observational cohort study examined the immediate effects of a customized interocclusal orthotic (CIO) on cervical muscle performance. Forty-two healthy adults (≥18 years) underwent strength and endurance testing with and without a CIO using a digital pressure gauge and six directional isometric contractions. Descriptive statistics and two-way repeated-measures MANOVA models were applied to evaluate condition effects. Results: CIO use produced significant improvements in cervical muscle strength and endurance across all directions compared to non-use. Forward flexion strength increased by 12.96% (p < 0.001, ηp² = 0.185), backward extension by 10.34% (p = 0.017, ηp² = 0.091), right rotation by 19.03% (p < 0.001, ηp² = 0.333) and left rotation by 19.86% (p < 0.001, ηp² = 0.353). Endurance gains demonstrated large effect sizes, with flexor endurance improving by 44.57% (p < 0.001, ηp² = 0.447). Conclusions: Optimized jaw alignment using a customized orthotic can elicit immediate, clinically meaningful enhancements in cervical strength and endurance, suggesting a promising adjunct for concussion risk mitigation in contact sports. Full article

Leveraging the complementary strengths of handcrafted radiomics and data-driven deep learning, this work develops and rigorously benchmarks three modeling streams (Models A, B and C) for pancreatic ductal adenocarcinoma (PDAC) detection on multiphase abdominal Computed Tomography (CT) scans. Model A distills hundreds of PyRadiomics descriptors to sixteen interpretable features that feed a gradient-boosted machine learning model, achieving discrimination (external AUC ≈ 0.99) with excellent calibration. Model B adopts a 3-D CBAM-ResNet-18 trained under weighted cross-entropy and mixed precision; although less accurate in isolation, it yields volumetric Grad-CAM maps that localize the tumor and provide explainability. Model C explores two fusion strategies that merge radiomics and deep embeddings: (i) a two-stage “frozen-stream” variant that locks both feature extractors and learns only a lightweight gating block plus classifier, and (ii) a full end-to-end version that allows the CNN’s adaptor layer to co-train with the fusion head. The frozen approach surpasses the single stream, whereas the end-to-end model reports external AUC of 0.987, balanced sensitivity/specificity above 0.93, and a Brier score below 0.05, while preserving clear Grad-CAM alignment with radiologist-drawn masks. Results demonstrate that a carefully engineered deep-radiomic fusion pipeline can deliver accurate, well-calibrated and interpretable PDAC triage directly from routine CT. Our contributions include a stability-verified 16-feature radiomic signature, a novel deep-radiomic fusion design that improves robustness and interpretability across vendors and a fully guideline-aligned, openly released pipeline for reproducible PDAC detection on routine CT. Full article

Accurate initial heading alignment is crucial for autonomous underwater vehicles (AUVs) relying on dead-reckoning (DR) navigation. The multiple GNSS position-based alignment (MGPA) method, using standard point positioning (SPP) GNSS, is an applicable approach in marine environments due to its standalone nature. However, the performance of this method is directly degraded by the inherent error in the initial position, which can be relatively large due to the use of SPP. Therefore, this paper proposes a novel iterative method that estimates and corrects errors in both initial heading and position. The core of the method is a decomposition of the coupled 2D optimization problem into two 1D optimizations by identifying an orthogonal correction basis. The effectiveness of the proposed method is validated through at-sea experiments with an AUV. Experimental results demonstrate that the proposed method corrects the initial position error and achieves improved alignment, enhancing DR navigation accuracy. Full article

Ship detection in remote sensing imagery is hindered by cluttered backgrounds, large variations in scale, and random orientations, limiting the performance of detectors designed for natural images. We propose RSDB-Net, a Rotation-Sensitive Dual-Branch Detection Network that introduces innovations in feature extraction, fusion, and detection. The Swin Transformer–CNN Backbone (STCBackbone) combines a Swin Transformer for global semantics with a CNN branch for local spatial detail, while the Feature Conversion and Coupling Module (FCCM) aligns and fuses heterogeneous features to handle multi-scale objects, and a Rotation-sensitive Cross-branch Fusion Head (RCFHead) enables bidirectional interaction between classification and localization, improving detection of randomly oriented targets. Additionally, an enhanced Feature Pyramid Network (eFPN) with learnable transposed convolutions restores semantic information while maintaining spatial alignment. Experiments on DOTA-v1.0 and HRSC2016 show that RSDB-Net performs better than the state of the art (SOTA), with mAP-ship values of 89.13% and 90.10% (+5.54% and +44.40% over the baseline, respectively), and reaches 72 FPS on an RTX 3090. RSDB-Net also demonstrates strong generalization and scalability, providing an effective solution for rotation-aware ship detection. Full article

Background/Objectives: Breast cancer diagnosis based on histopathological examination remains the most reliable and widely accepted approach in clinical practice, despite being time-consuming and prone to inter-observer variability. While deep learning methods have achieved high accuracy in medical image classification, their cross-site generalization remains limited due to differences in staining protocols and image acquisition. This study aims to evaluate and compare three clinically relevant adaptation strategies to improve model robustness under domain shift. Methods: The ResNet50V2 model, pretrained on ImageNet and further fine-tuned on the Kaggle Breast Histopathology Images dataset, was subsequently adapted to the BreaKHis dataset under three clinically relevant transfer strategies: (i) threshold calibration without retraining (site calibration), (ii) head-only fine-tuning (light FT), and (iii) full fine-tuning (full FT). Experiments were performed on an internal balanced dataset and on the public BreaKHis dataset using strict patient-level splitting to avoid data leakage. Evaluation metrics included accuracy, precision, recall, F1-score, ROC-AUC, and PR-AUC, computed per magnification level (40×, 100×, 200×, 400×). Results: Full fine-tuning consistently yielded the highest performance across all magnifications, reaching up to 0.983 ROC-AUC and 0.980 sensitivity at 400×. At 40× and 100×, the model correctly identified over 90% of malignant cases, with ROC-AUC values of 0.9500 and 0.9332, respectively. Head-only fine-tuning led to moderate gains (e.g., sensitivity up to 0.859 at 200×), while threshold calibration showed limited improvements (ROC-AUC ranging between 0.60–0.73). Grad-CAM analysis revealed more stable and focused attention maps after full fine-tuning, though they did not always align with diagnostically relevant regions. Conclusions: Our findings confirm that full fine-tuning is essential for robust cross-site deployment of histopathology AI systems, particularly at high magnifications. Lighter strategies such as threshold calibration or head-only fine-tuning may serve as practical alternatives in resource-constrained environments where retraining is not feasible. Full article

Along with the rapid advancement of Virtual Reality (VR) and the metaverse, interest in this technology has surged among game developers and in fields such as education and healthcare. VR has enabled the rise in immersive, gamified activities, whether for rehabilitation, therapy, or learning. Additionally, VR and Motion Capture (MoCap) have allowed developers to create further accessibility features for end-users with special needs. However, the excitement of using new technology often does not align with the end user’s use cases. The over-reliance on cutting-edge hardware can negatively impact most end users who lack access to such expensive tools. To this end, we conducted an inclusivity-focused study that enables learners to practice ASL in an immersive and engaging way using only head- and controller-based tracking. Our approach replaces full-body MoCap with Inverse Kinematics (IK) and simple controller mappings for upper-body pose and hand-gesture recognition, providing a low-cost, reproducible alternative to costly setups. Full article

Background: Despite evidence of potential benefits, variability in exercise types, psychological outcomes, and methods hinders comprehensive cost-effectiveness evaluation, framed through Stimulus–Organism–Response (S–O–R) theory. In this context, Workplace Physical Activity-Based Programs (WPABPs) serve as environmental stimulation that influences employees’ emotional states, which in turn shape mental health outcomes and behavioral responses. Research Purpose: This systematic review examines WPABPs through the social dimension of the Environmental, Social, Governance (ESG-S) framework, with a focus on their impact on employees’ mental health. Methods: Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines, eligibility was assessed via the PICO (Population, Intervention, Comparison, Outcome) framework. The ScienceDirect, Scopus, Google Scholar, and PubMed databases were searched using Medical Subject Headings (MeSH) aligned keywords and Boolean operators. Results: Of the 961 articles identified, 15 studies (2021–2025) met the inclusion criteria. WPABPs were found to improve employee mental health, reduce stress, and enhance well-being. Individualized interventions supported targeted psychological benefits, while group formats promoted social cohesion and engagement. Variations in type, duration, and delivery, as well as accessibility barriers for underrepresented employees, were noted. WPABPs enhance employee well-being and organizational outcomes, contributing to the Sustainable Development Goals (SDGs), specifically SDG 3 (Good Health and Well-being) and SDG 8 (Decent Work and Economic Growth). Conclusions: Hybrid models combining individual and group approaches with managerial and digital support are recommended. Integrating WPABPs within ESG-S and Corporate Social Responsibility (CSR) frameworks can institutionalize sustainable workplace health promotion, while future research should focus on standardized, inclusive, and long-term evaluations. Full article

Traditional calibration methods rely on precise targets and frequent manual intervention, making them time-consuming and unsuitable for large-scale deployment. Existing learning-based approaches, while automating the process, are typically limited to single LiDAR–camera pairs, resulting in poor scalability and high computational overhead. To address these limitations, we propose a lightweight calibration network with flexibility in the number of sensor pairs, making it capable of jointly calibrating multiple cameras and LiDARs in a single forward pass. Our method employs a frozen pre-trained Swin Transformer as a shared backbone to extract unified features from both RGB images and corresponding depth maps. Additionally, we introduce a cross-modal channel-wise attention module to enhance key feature alignment and suppress irrelevant noise. Moreover, to handle variations in viewpoint, we design a modular calibration head that independently estimates the extrinsics for each LiDAR–camera pair. Through large-scale experiments on the nuScenes dataset, we show that our model, requiring merely 78.79 M parameters, attains a mean translation error of 2.651 cm and a rotation error of $0.{246}^{\circ }$, achieving comparable performance to existing methods while significantly reducing the computational cost. Full article

The accurate identification of soybean growth stages is critical for optimizing agricultural interventions, where mistimed treatments can result in yield losses ranging from 2.5% to 40%. Existing deep learning approaches remain limited in scope, targeting isolated developmental phases rather than providing comprehensive phenological coverage. This paper presents a novel object detection architecture DELTA-SoyStage, combining an EfficientNet backbone with a lightweight ChannelMapper neck and a newly proposed DELTA (Denoising Enhanced Lightweight Task Alignment) detection head for soybean growth stage classification. We introduce a dataset of 17,204 labeled RGB images spanning nine growth stages from emergence (VE) through full maturity (R8), collected under controlled greenhouse conditions with diverse imaging angles and lighting variations. DELTA-SoyStage achieves 73.9% average precision with only 24.4 GFLOPs computational cost, demonstrating 4.2× fewer FLOPs than the best-performing baseline (DINO-Swin: 74.7% AP, 102.5 GFLOPs) with only 0.8% accuracy difference. The lightweight DELTA head combined with the efficient ChannelMapper neck requires only 8.3 M parameters—a 43.5% reduction compared to standard architectures—while maintaining competitive accuracy. Extensive ablation studies validate key design choices including task alignment mechanisms, multi-scale feature extraction strategies, and encoder–decoder depth configurations. The proposed model’s computational efficiency makes it suitable for deployment on resource-constrained edge devices in precision agriculture applications, enabling timely decision-making without reliance on cloud infrastructure. Full article

Citrus Huanglongbing (HLB) is one of the most destructive diseases in the global citrus industry; its pathogen is transmitted primarily by the Asian citrus psyllid (ACP), Diaphorina citri Kuwayama, making timely monitoring and control of ACP populations essential. Real-world ACP monitoring faces several challenges, including tiny targets easily confused with the background, noise amplification and spurious detections caused by textures, stains, and specular glare on yellow-boards, unstable localization due to minute shifts of small boxes, and strict constraints on parameters, computation, and model size for long-term edge deployment. To address these challenges, we focus on the yellow-board ACP monitoring scenario and create the ACP Yellow Sticky Trap Dataset (ACP-YSTD), which standardizes background and acquisition procedures, covering common interference sources. The dataset consists of 600 images with 3837 annotated ACP, serving as a unified basis for training and evaluation. On the modeling side, we propose TGSP-YOLO11, an improved YOLO11-based detector: the detection head is reconfigured to the two scales P2 + P3 to match tiny targets and reduce redundant paths; Guided Scalar Fusion (GSF) is introduced on the high-resolution branch to perform constrained, lightweight scalar fusion that suppresses noise amplification; ShapeIoU is adopted for bounding-box regression to enhance shape characterization and alignment robustness for small objects; and Network Slimming is employed for channel-level structured pruning, markedly reducing parameters, FLOPs, and model size to satisfy edge deployment, without degrading detection performance. Experiments show that on the ACP-YSTD test set, TGSP-YOLO11 achieves precision 92.4%, recall 95.5%, and F1 93.9, with 392,591 parameters, a model size of 1.4 MB, and 6.0 GFLOPs; relative to YOLO11n, recall increases by 4.6%, F1 by 2.4, and precision by 0.2%, while the parameter count, model size, and computation decrease by 84.8%, 74.5%, and 4.8%, respectively. Compared to representative detectors (SSD, RT-DETR, YOLOv7-tiny, YOLOv8n, YOLOv9-tiny, YOLOv10n, YOLOv12n, YOLOv13n), TGSP-YOLO11 improves recall by 33.9%, 19.0%, 8.5%, 10.1%, 6.3%, 4.6%, 6.9%, and 5.7%, respectively, and F1 by 19.9, 14.9, 5.1, 6.0, 2.6, 5.6, 3.6, and 3.9, respectively. Additionally, it reduces parameter count, model size, and computation by 84.0–98.8%, 74.5–97.9%, and 3.2–94.2%, respectively. Transfer evaluation indicates that on 20 independent yellow-board images not seen during training, the model attains precision 94.3%, recall 95.8%, F1 95.0, and 159.2 FPS. Full article

Federated learning (FL) for skeleton-based action recognition remains underexplored, particularly under strong client heterogeneity where regular FedAvg tends to cause client drift and unstable convergence. We introduce Clustered Federated Spatio-Temporal Graph Attention Networks (CF-STGAT), a clustered FL framework that leverages attention-derived spatio-temporal statistics from local STGAT models to dynamically group clients and perform attention-weighted inter-cluster fusion that gently align cluster models. Concretely, the server periodically extracts multi-head parameter-based attention descriptors, normalizes and projects them via PCA, and applies K-means to form clusters; a global reference is then computed by attention–similarity weighting and used to regularize each cluster model with a lightweight fusion step. On NTU RGB+D 60/120(NTU 60/120), CF-STGAT consistently outperforms strong FL baselines with the STGAT backbone, yielding absolute top-1 gains of +0.84/+4.09 (NTU 60, X-Sub/X-Setup) and +7.98/+4.18 (NTU 120, X-Sub/X-Setup) over FedAvg, alongside smoother per-client trajectories and lower terminal test loss. Ablations indicate that attention-guided clustering and inter-cluster fusion are complementary: clustering reduces within-group variance whereas fusion limits cross-cluster divergence. The approach keeps local training unchanged and adds only server-side statistics and clustering. Full article

Background: Breast cancer is a leading cause of cancer-related mortality among women, and earlier diagnosis significantly improves treatment outcomes. However, traditional mammography-based systems rely on single-modality image analysis and lack integration of volumetric and clinical context, which limits diagnostic robustness. Deep learning models have shown promising results in identification but are typically restricted to 2D feature extraction and lack cross-modal reasoning capability. Objective: This study proposes SwinCAMF-Net, a multimodal cross-attention Swin transformer network designed to improve joint breast lesion classification and segmentation by integrating multi-view mammography, 3D ROI volumes, and clinical metadata. Methods: SwinCAMF-Net employs a Swin transformer encoder for hierarchical visual representation learning from mammographic views, a 3D CNN volume encoder for lesion depth context modelling, and a clinical projection module to embed patient metadata. A novel cross-attentive fusion (CAF) module selectively aligns multimodal features through query–key attention. The fused feature representation branches into a classification head for malignancy prediction and a segmentation decoder for lesion localization. The model is trained and evaluated on CBIS-DDSM and RTM benchmark datasets. Results: SwinCAMF-Net achieved accuracy up to 0.978, an AUC-ROC of 0.998, and an F1-score of 0.944 for classification, while segmentation reached a Dice coefficient of 0.931. Ablation experiments confirm that the CAF module improves performance by up to 6.9%, demonstrating its effectiveness in multimodal fusion. Conclusion: SwinCAMF-Net advances breast cancer analysis by providing complementary multimodal evidence through a cross-attentive fusion, leading to improved diagnostic performance and clinical interpretability. The framework demonstrates strong potential in AI-assisted screening and radiology decision support. Full article

Background/Objectives: Subcortical local field potentials (LFPs) provide a valuable in vivo window into the neurophysiology of the dystonia network. These signals can be recorded through Deep Brain Stimulation (DBS) devices and combined with whole-head techniques such as magnetoencephalography (MEG) to study cortical–subcortical interactions. However, simultaneous LFP-MEG acquisition poses challenges, including interference from the DBS device and synchronization issues. We present preliminary data on the feasibility and signal quality of concurrent LFP and MEG recordings in dystonia patients. Methods: We assessed simultaneous MEG-LFP recordings in 11 patients with inherited or idiopathic dystonia who underwent bilateral DBS lead implantation in the Globus Pallidus Internus (GPi). Two synchronization strategies were tested: (1) the Tapping method, using an accelerometer placed on the DBS device, and (2) the Stimulation method, which generated detectable artifacts during sham stimulation. Results: Both methods successfully aligned MEG and LFP signals with a mean temporal delay of 91 ± 22 ms for the Tapping method and 288 ± 166 ms for the Stimulation method. Post-implantation signal-to-noise ratio analysis revealed slight degradation but no significant impact on MEG quality (gradiometers: −0.12 ± 1.85 dB; magnetometers: −0.47 ± 2.03 dB). Conclusions: Simultaneous MEG-LFP recordings in dystonic patients are feasible, yielding high-quality signals, and reliable synchronization. Temporal alignment improved with practice, suggesting a short learning curve. This method opens new opportunities to study cortical-subcortical dynamics and strengthens the potential of combining MEG-LFP approaches for investigating dystonia. Full article

Unmanned aerial vehicles (UAVs) have been widely used in aerial photography and target detection tasks due to their flexibility and unique perspective. However, small targets often suffer from insufficient resolution, uneven scale distribution, and complex background clutter, which are constrained by imaging conditions such as high-altitude imaging, long-distance capture, and wide field of view. These factors weaken the feature representation and generalization ability of the model, becoming the key bottleneck that restricts the improvement of small target detection accuracy in UAV scenarios. To address the above issues, this paper proposes a small target detection algorithm for UAV perspective, namely MTD-YOLO. First, a Parallel Multi-Scale Receptive Field Unit (PMSRFU) is designed. This unit effectively enhances the receptive field range of feature extraction and the fusion ability of multi-scale contextual information by introducing parallel branches with different-sized convolutional kernels. Second, we embed PMSRFU into a C2f block to form C2f-PMSRFU, which reuses shallow details and fuses multi-scale features to clarify edges and textures in small targets, yielding stronger fine-grained representations. Finally, an efficient detection head with task decoupling, dynamic alignment, and adaptive scale adjustment capabilities, namely SDIDA-Head, is proposed, which significantly improves the model’s small target detection accuracy. Extensive experiments on the VisDrone2019 and HazyDet datasets demonstrate that MTD-YOLO achieves a 7.6% and 6.6% increase in mAP@0.5 compared to the baseline YOLOv8n, respectively. Meanwhile, the Precision is improved by 6.0% and 1.1%, and the Recall is enhanced by 7.5% and 6.9%, respectively. These results fully validate the effectiveness and superiority of the proposed method in UAV small target detection tasks. Full article