Search Results (2,379)

Purpose: Luggage X-rays suffer from low contrast, material overlap, and noise; dual-energy imaging reduces ambiguity but creates colour biases that impair segmentation. This study aimed to (1) employ connotative fusion by embedding realistic detonator patches into real X-rays to simulate threats and enhance unattended detection without requiring ground-truth labels; (2) thoroughly evaluate fusion techniques in terms of balancing image quality, information content, contrast, and the preservation of meaningful features. Methods: A total of 1000 X-ray luggage images and 150 detonator images were used for fusion experiments based on deep learning, transform-based, and feature-driven methods. The proposed approach does not need ground truth supervision. Deep learning fusion techniques, including VGG, FusionNet, and AttentionFuse, enable the dynamic selection and combination of features from multiple input images. The transform-based fusion methods convert input images into different domains using mathematical transforms to enhance fine structures. The Nonsubsampled Contourlet Transform (NSCT), Curvelet Transform, and Laplacian Pyramid (LP) are employed. Feature-driven image fusion methods combine meaningful representations for easier interpretation. Singular Value Decomposition (SVD), Principal Component Analysis (PCA), Random Forest (RF), and Local Binary Pattern (LBP) are used to capture and compare texture details across source images. Entropy (EN), Standard Deviation (SD), and Average Gradient (AG) assess factors such as spatial resolution, contrast preservation, and information retention and are used to evaluate the performance of the analysed methods. Results: The results highlight the strengths and limitations of the evaluated techniques, demonstrating their effectiveness in producing sharpened fused X-ray images with clearly emphasized targets and enhanced structural details. Conclusions: The Laplacian Pyramid fusion method emerges as the most versatile choice for applications demanding a balanced trade-off. This is evidenced by its overall multi-criteria balance, supported by a composite (geometric mean) score on normalised metrics. It consistently achieves high performance across all evaluated metrics, making it reliable for detecting concealed threats under diverse imaging conditions. Full article

Timely identification and accurate delineation of ultra-early ischemic stroke lesions in non-contrast computed tomography (CT) scans of the human brain are of paramount importance for prompt medical intervention and improved patient outcomes. In this study, we propose a deep learning-driven methodology specifically designed for segmenting ultra-early ischemic regions, with a particular emphasis on both the ischemic core and the surrounding penumbra during the initial stages of stroke progression. We introduce a lightweight preprocessing model based on convolutional filtering techniques, which enhances image clarity while preserving the structural integrity of medical scans, a critical factor when detecting subtle signs of ultra-early ischemic strokes. Unlike conventional preprocessing methods that directly modify the image and may introduce artifacts or distortions, our approach ensures the absence of neural network-induced artifacts, which is especially crucial for accurate diagnosis and segmentation of ultra-early ischemic lesions. The model employs predefined differentiable filters with trainable parameters, allowing for artifact-free and precision-enhanced image refinement tailored to the challenges of ultra-early stroke detection. In addition, we incorporated into the combined preprocessing pipeline a newly proposed trainable linear combination of pretrained image filters, a concept first introduced in this study. For model training and evaluation, we utilize a publicly available dataset of acute ischemic stroke cases, focusing on the subset relevant to ultra-early stroke manifestations, which contains annotated non-contrast CT brain scans from 112 patients. The proposed model demonstrates high segmentation accuracy for ultra-early ischemic regions, surpassing existing methodologies across key performance metrics. The results have been rigorously validated on test subsets from the dataset, confirming the effectiveness of our approach in supporting the early-stage diagnosis and treatment planning for ultra-early ischemic strokes. Full article

Breast cancer is the most commonly diagnosed cancer in women globally and represents the leading cause of mortality related to malignant tumors. Currently, healthcare professionals are focused on developing and implementing innovative techniques to improve the early detection of this disease. Thermography, studied as a complementary method to traditional approaches, captures infrared radiation emitted by tissues and converts it into data about skin surface temperature. During tumor development, angiogenesis occurs, increasing blood flow to support tumor growth, which raises the surface temperature in the affected area. Automatic classification techniques have been explored to analyze thermographic images and develop an optimal classification tool to identify thermal anomalies. This study aims to design a concise description using statistical and texture features to accurately classify thermograms as control or highly probable to be cancer (with thermal anomalies). The importance of employing a short description lies in facilitating interpretation by medical professionals. In contrast, a characterization based on a large number of variables could make it more challenging to identify which values differentiate the thermograms between groups, thereby complicating the explanation of results to patients. A maximum accuracy of 91.97% was achieved by applying only seven features and using a Coarse Decision Tree (DT) classifier and robust Machine Learning (ML) model, which demonstrated competitive performance compared with previously reported studies. Full article

Background: Accurate prediction of postoperative care requirements is critical for patient safety and resource allocation. Although numerous approaches involving artificial intelligence (AI) and machine learning (ML) have been proposed to support such predictions, their implementation in practice has so far been insufficiently successful. One reason for this is that the performance of the algorithms is difficult to assess in practical use, as the accuracy of clinical decisions has not yet been systematically quantified. As a result, models are often assessed purely from a technical perspective, neglecting the socio-technical context. Methods: We conducted a retrospective, single-center observational study at the University Hospital Augsburg, including 35,488 elective surgical cases documented between August 2023 and January 2025. For each case, preoperative care-level predictions by surgical and anesthesiology teams were compared with the actual postoperative care provided. Predictive performance was evaluated using accuracy and sensitivity. Since this is a highly imbalanced dataset, in addition to sensitivity and specificity, the balanced accuracy and the $F_{\beta }$-score were also calculated. The results were contrasted with published Machine-Learning (ML)-based approaches. Results: Overall prediction accuracy was high (surgery: 91.2%; anesthesiology: 87.1%). However, sensitivity for identifying patients requiring postoperative intensive care was markedly lower than reported for ML models in the literature, with the largest discrepancies observed in patients ultimately admitted to the ICU (surgery: 38.05%; anesthesiology: 56.84%; ML: 70%). Nevertheless, clinical judgment demonstrated a superior $F_1$-score, indicating a more balanced performance between sensitivity and precision (surgery: 0.527; anesthesiology: 0.551; ML: 0.28). Conclusions: This study provides the first real-world benchmark of clinical expertise in postoperative care prediction and shows a way in which modern ML approaches must be evaluated in a specific sociotechnical context. By quantifying the predictive performance of surgeons and anesthesiologists, it enables an evaluation of existing ML approaches. Thus the strength of our work is the provision of a real-world benchmark against which all ML methods for preoperative prediction of ICU demand can be systematically evaluated. This enables, for the first time, a comparison of different approaches on a common, practice-oriented basis and thus significantly facilitates translation into clinical practice, thereby closing the translational gap. Furthermore it offers a data-driven framework to support the integration of ML into preoperative decision-making. Full article

Teacher preparation programmes must now ensure instructional continuity and quality across face-to-face, online, and hybrid modes, even amid health, climate, or security crises. This mixed-methods study examined which principles policymakers and teacher education directors deem essential for such resilience, and how those principles align with prior research and leadership theory. Semi-structured elite interviews (N = 25) were analyzed inductively to surface field-driven themes and deductively through two models: the ten evidence-based training principles synthesized by Hadad et al. and the six capacities of Striepe and Cunningham’s Crises Leadership Framework (CLF). Results show strong consensus on theory–practice integration, university–school partnerships, and collaborative learning, mapping chiefly to the CLF capacities of adaptive roles and stakeholder collaboration. Directors added practice-oriented priorities—authentic field immersion, formative feedback, and inclusive pedagogy—extending the crisis care and contextual influence dimensions. By contrast, policymakers uniquely stressed policy–academic co-decision-making, reinforcing complex decision-making at the system level. Reflective thinking skills and digital pedagogy, though prominent in the literature, were under-represented, signalling implementation gaps. Overall, the integrated model offers a crisis-ready blueprint for curriculum design, partnership governance, and digital capacity-building that can sustain continuity and quality in pre-service teacher education. Full article

The filarioid nematode Dirofilaria immitis is the causative agent of heartworm disease, a major parasitic infection of canids, felids and occasionally humans. Current prevention relies on macrocyclic lactone-based chemoprophylaxis, but the emergence of drug resistance highlights the need for new intervention strategies. Here, we applied a machine learning (ML)-based framework to predict and prioritise essential genes in D. immitis in silico, using genomic, transcriptomic and functional datasets from the model organisms Caenorhabditis elegans and Drosophila melanogaster. With a curated set of 26 predictive features, we trained and evaluated multiple ML models and, using a defined threshold, we predicted 406 ‘high-priority’ essential genes. These genes showed strong transcriptional activity across developmental stages and were inferred to be enriched in pathways related to ribosome biogenesis, translation, RNA processing and signalling, underscoring their potential as anthelmintic targets. Transcriptomic analyses suggested that these genes are associated with key reproductive and neural tissues, while chromosomal mapping revealed a relatively even genomic distribution, in contrast to patterns observed in C. elegans and Dr. melanogaster. In addition, initial evidence suggested structural variation in the X chromosome compared with a recently published D. immitis assembly, indicating the importance of integrating long-read sequencing with high-throughput chromosome conformation capture (Hi-C) mapping. Overall, this study reinforces the potential of ML-guided approaches for essential gene discovery in parasitic nematodes and provides a foundation for downstream validation and therapeutic target development. Full article

We present a visual–inertial odometry (VIO) system that integrates a deep feature extraction and filtering strategy with optical flow to improve tracking robustness. While many traditional VIO methods rely on hand-crafted features, they often struggle to remain robust under challenging visual conditions, such as low texture, motion blur, or lighting variation. These methods tend to exhibit large performance variance across different environments, primarily due to the limited repeatability and adaptability of hand-crafted keypoints. In contrast, learning-based features offer richer representations and can generalize across diverse domains thanks to data-driven training. However, they often suffer from uneven spatial distribution and temporal instability, which can degrade tracking performance. To address these issues, we propose a hybrid front-end that combines a lightweight deep feature extractor with an image pyramid and grid-based keypoint sampling to enhance spatial diversity. Additionally, a forward–backward optical-flow-consistency check is applied to filter unstable keypoints. The system improves feature tracking stability by enforcing spatial and temporal consistency while maintaining real-time efficiency. Finally, the effectiveness of the proposed VIO system is validated on the EuRoC MAV benchmark, showing a 19.35% reduction in trajectory RMSE and improved consistency across multiple sequences compared with previous methods. Full article

Decentralized energy trading has been designed as a scalable substitute for traditional electricity markets. While blockchain technology facilitates efficient transparency and automation for peer-to-peer energy trading, the majority of current proposals lack real-time intelligence and adaptability concerning pricing strategies. This paper presents an innovative machine learning-driven solar energy trading platform on the Ethereum blockchain that uniquely integrates Bayesian-optimized XGBoost models with dynamic pricing mechanisms inherently incorporated within smart contracts. The principal innovation resides in the real-time amalgamation of meteorological data via Chainlink oracles with machine learning-enhanced price optimization, thereby establishing an adaptive system that autonomously responds to fluctuations in supply and demand. In contrast to existing static pricing methodologies, our framework introduces a multi-faceted dynamic pricing model that encompasses peak-hour adjustments, prediction confidence weighting, and weather-influenced corrections. The system dynamically establishes energy prices predicated on real-time supply–demand forecasts through the implementation of role-based access control, cryptographic hash functions, and ongoing integration of meteorological and machine learning data. Utilizing real-world meteorological data from La Trobe University’s UNISOLAR dataset, the Bayesian-optimized XGBoost model attains a remarkable prediction accuracy of 97.45% while facilitating low-latency price updates at 30 min intervals. The proposed system delivers robust transaction validation, secure offer creation, and scalable dynamic pricing through the seamless amalgamation of off-chain machine learning inference with on-chain smart contract execution, thereby providing a validated platform for trustless, real-time, and intelligent decentralized energy markets that effectively address the disparity between theoretical blockchain energy trading and practical implementation needs. Full article

Understanding the intricate relationship between composition, processing conditions, and material properties is essential for optimizing Cu-based alloys. Machine learning offers a powerful tool for decoding these complex interactions, enabling more efficient alloy design. This work introduces a comprehensive machine learning framework aimed at accurately predicting key properties such as hardness and electrical conductivity of low-alloyed Cu-based alloys. By integrating various input parameters, including chemical composition and thermo-mechanical processing parameters, the study develops and validates multiple machine learning models, including Multi-Layer Perceptron with Production-Aware Deep Architecture (MLP-PADA), Deep Feedforward Network with Multi-Regularization Framework (DFF-MRF), Feedforward Network with Self-Adaptive Optimization (FFN-SAO), and Feedforward Network with Materials Mapping (FFN-TMM). On a held-out test set, DFF-MRF achieved the best generalization (R²_test = 0.9066; RMSE_test = 5.3644), followed by MLP-PADA (R²_test = 0.8953; RMSE_test = 5.7080) and FFN-TMM (R²_test = 0.8914; RMSE_test = 5.8126), with FFN-SAO slightly lower (R²_test = 0.8709). Additionally, a computational performance analysis was conducted to evaluate inference time, memory usage, energy consumption, and batch scalability across all models. Feature importance analysis was conducted, revealing that aging temperature, Cr, and aging duration were the most influential factors for hardness. In contrast, aging duration, aging temperature, solution treatment temperature, and Cu played key roles in electrical conductivity. The results demonstrate the effectiveness of these advanced machine learning models in predicting critical material properties, offering insightful advancements for materials science research. This study introduces the first controlled, statistically validated, multi-model benchmark that integrates composition and thermo-mechanical processing with deployment-grade profiling for property prediction of low-alloyed Cu alloys. Full article

As network automation and self-organizing networks (SONs) rapidly evolve, edge devices increasingly demand lightweight, real-time, and high-precision classification algorithms to support critical tasks such as traffic identification, intrusion detection, and fault diagnosis. In recent years, cross-entropy (CE) loss has been widely adopted in deep learning classification tasks due to its computational efficiency and ease of optimization. However, traditional CE methods primarily focus on class separability without explicitly constraining intra-class compactness and inter-class boundaries in the feature space, thereby limiting their generalization performance on complex classification tasks. To address this issue, we propose a novel classification loss framework—Contrastive Geometric Cross-Entropy (CGCE). Without incurring additional computational or memory overhead, CGCE explicitly introduces learnable class representation vectors and constructs the loss function based on the dot-product similarity between features and these class representations, thus explicitly reinforcing geometric constraints in the feature space. This mechanism effectively enhances intra-class compactness and inter-class separability. Theoretical analysis further demonstrates that minimizing the CGCE loss naturally induces clear and measurable geometric class boundaries in the feature space, a desirable property absent from traditional CE methods. Furthermore, CGCE can seamlessly incorporate the prior knowledge of pretrained models, converging rapidly within only a few training epochs (for example, on the CIFAR-10 dataset using the ViT model, a single training epoch is sufficient to reach 99% of the final training accuracy.) Experimental results on both text and image classification tasks show that CGCE achieves accuracy improvements of up to 2% over traditional CE methods, exhibiting stronger generalization capabilities under challenging scenarios such as class imbalance, few-shot learning, and noisy labels. These findings indicate that CGCE has significant potential as a superior alternative to traditional CE methods. Full article

As cities strive for human-centered and fine-tuned development, Agent-Based Modeling (ABM) has emerged as a powerful tool for simulating pedestrian behavior and optimizing walkable neighborhood design. This study presents a comparative bibliometric analysis of ABM applications in block-scale walkability research from 2015 to 2024, drawing on both Chinese- and English-language literature. Using visualization tools such as VOSviewer, the analysis reveals divergences in national trajectories, methodological approaches, and institutional logics. Chinese research demonstrates a policy-driven growth pattern, particularly following the introduction of the “15-Minute Community Life Circle” initiative, with an emphasis on neighborhood renewal, age-friendly design, and transit-oriented planning. In contrast, international studies show a steady output driven by technological innovation, integrating methods such as deep learning, semantic segmentation, and behavioral simulation to address climate resilience, equity, and mobility complexity. The study also classifies ABM applications into five key application domains, highlighting how Chinese and international studies differ in focus, data inputs, and implementation strategies. Despite these differences, both research streams recognize the value of ABM in transport planning, public health, and low-carbon urbanism. Key challenges identified include data scarcity, algorithmic limitations, and ethical concerns. The study concludes with future research directions, including multimodal data fusion, integration with extended reality, and the development of privacy-aware, cross-cultural modeling standards. These findings reinforce ABM’s potential as a smart urban simulation tool for advancing adaptive, human-centered, and sustainable neighborhood planning. Full article

Open-vocabulary object detection in remote sensing aims to detect novel categories not seen during training, which is crucial for practical aerial image analysis applications. While some approaches accomplish this task through large-scale data construction, such methods incur substantial annotation and computational costs. In contrast, we focus on efficient utilization of limited datasets. However, existing methods such as CastDet struggle with inefficient data utilization and class imbalance issues in pseudo-label generation for novel categories. We propose an enhanced open-vocabulary detection framework that addresses these limitations through two key innovations. First, we introduce a selective masking strategy that enables direct utilization of partially annotated images by masking base category regions in teacher model inputs. This approach eliminates the need for strict data separation and significantly improves data efficiency. Second, we develop a dynamic frequency-based class weighting that automatically adjusts category weights based on real-time pseudo-label statistics to mitigate class imbalance issues. Our approach integrates these components into a student–teacher learning framework with RemoteCLIP for novel category classification. Comprehensive experiments demonstrate significant improvements on both datasets: on VisDroneZSD, we achieve 42.7% overall mAP and 41.4% harmonic mean, substantially outperforming existing methods. On DIOR dataset, our method achieves 63.7% overall mAP with 49.5% harmonic mean. Our framework achieves more balanced performance between base and novel categories, providing a practical and data-efficient solution for open-vocabulary aerial object detection. Full article

The background of the article refers to the diagnosis of focal liver lesions (FLLs) through contrast-enhanced ultrasound (CEUS) based on the integration of spatial and temporal information. Traditional computer-aided diagnosis (CAD) systems predominantly rely on static images, which limits the characterization of lesion dynamics. This study aims to assess the effectiveness of Transformer-based architectures in enhancing CAD performance within the realm of liver pathology. The methodology involved a systematic comparison of deep learning models for the analysis of CEUS images and videos. For image-based classification, a Hybrid Transformer Neural Network (HTNN) was employed. It combines Vision Transformer (ViT) modules with lightweight convolutional features. For video-based tasks, we evaluated a custom spatio-temporal Convolutional Neural Network (CNN), a CNN with Long Short-Term Memory (LSTM), and a Video Vision Transformer (ViViT). The experimental results show that the HTNN achieved an outstanding accuracy of 97.77% in classifying various types of FLLs, although it required manual selection of the region of interest (ROI). The video-based models produced accuracies of 83%, 88%, and 88%, respectively, without the need for ROI selection. In conclusion, the findings indicate that Transformer-based models exhibit high accuracy in CEUS-based liver diagnosis. This study highlights the potential of attention mechanisms to identify subtle inter-class differences, thereby reducing the reliance on manual intervention. Full article

Single-object vision–language tracking has become an important research topic due to its potential in applications such as intelligent surveillance and autonomous driving. However, existing cross-modal alignment methods typically rely on contrastive learning and struggle to effectively address semantic ambiguity or the presence of multiple similar objects. This study aims to explore how to achieve more robust vision–language alignment under these challenging conditions, thereby achieving accurate object localization. To this end, we propose a text heatmap mapping (THM) module that enhances the spatial guidance of textual cues in tracking. The THM module integrates visual and language features and generates semantically aware heatmaps, enabling the tracker to focus on the most relevant regions while suppressing distractors. This framework, developed based on UVLTrack, combines a visual transformer with a pre-trained language encoder. The proposed method is evaluated on benchmark datasets such as OTB99, LaSOT, and TNL2K. The main contribution of this paper is the introduction of a novel spatial alignment mechanism for multimodal tracking and its effectiveness on various tracking benchmarks. Results demonstrate that the THM-based tracker improves robustness to semantic ambiguity and multi-instance interference, outperforming baseline frameworks. Full article

Hyperedge prediction is crucial for uncovering higher-order relationships in complex systems but faces core challenges, including unmodeled node influence heterogeneity, overlooked hyperedge order effects, and data sparsity. This paper proposes Order propagation Fusion Self-supervised learning for Hyperedge prediction (OFSH) to address these issues. OFSH introduces a hyperedge order propagation mechanism that dynamically learns node importance weights and groups neighbor hyperedges by order, applying max–min pooling to amplify feature distinctions. To mitigate data sparsity, OFSH incorporates a key node-guided augmentation strategy with adaptive masking, preserving core high-order semantics. It identifies topological hub nodes based on their comprehensive influence and employs adaptive masking probabilities to generate augmented views preserving core high-order semantics. Finally, a triadic contrastive loss is employed to maximize cross-view consistency and capture invariant semantic information under perturbations. Extensive experiments on five public real-world hypergraph datasets demonstrate significant improvements over state-of-the-art methods in AUROC and AP. Full article