Search Results (17,198)

The retina offers a unique window into both ocular and systemic health, motivating the development of AI-based tools for disease screening and risk assessment. In this study, we present a comprehensive evaluation of six state-of-the-art deep neural networks, including convolutional neural networks and vision transformer architectures, on the Brazilian Multilabel Ophthalmological Dataset (BRSET), comprising 16,266 fundus images annotated for multiple clinical and demographic labels. We explored seven classification tasks: Diabetes, Diabetic Retinopathy (2-class), Diabetic Retinopathy (3-class), Hypertension, Hypertensive Retinopathy, Drusen, and Sex classification. Models were evaluated using precision, recall, F1-score, accuracy, and AUC. Among all models, the Swin-L generally delivered the best performance across scenarios for Diabetes (AUC = 0.88, weighted F1-score = 0.86), Diabetic Retinopathy (2-class) (AUC = 0.98, weighted F1-score = 0.95), Diabetic Retinopathy (3-class) (macro AUC = 0.98, weighted F1-score = 0.95), Hypertension (AUC = 0.85, weighted F1-score = 0.79), Hypertensive Retinopathy (AUC = 0.81, weighted F1-score = 0.97), Drusen detection (AUC = 0.93, weighted F1-score = 0.90), and Sex classification (AUC = 0.87, weighted F1-score = 0.80). These results reflect excellent to outstanding diagnostic performance. We also employed gradient-based saliency maps to enhance explainability and visualize decision-relevant retinal features. Our findings underscore the potential of deep learning, particularly vision transformer models, to deliver accurate, interpretable, and clinically meaningful screening tools for retinal and systemic disease detection. Full article

The classification of submarine topography and geomorphology is essential for marine resource exploitation and ocean engineering, with wide-ranging implications in marine geology, disaster assessment, resource exploration, and autonomous underwater navigation. Submarine landscapes are highly complex and diverse. Traditional visual interpretation methods are not only inefficient and subjective but also lack the precision required for high-accuracy classification. While many machine learning and deep learning models have achieved promising results in image classification, limited work has been performed on integrating backscatter and bathymetric data for multi-source processing. Existing approaches often suffer from high computational costs and excessive hyperparameter demands. In this study, we propose a novel approach that integrates pruning-enhanced ConDenseNet with label smoothing regularization to reduce misclassification, strengthen the cross-entropy loss function, and significantly lower model complexity. Our method improves classification accuracy by 2% to 10%, reduces the number of hyperparameters by 50% to 96%, and cuts computation time by 50% to 85.5% compared to state-of-the-art models, including AlexNet, VGG, ResNet, and Vision Transformer. These results demonstrate the effectiveness and efficiency of our model for multi-source submarine topography classification. Full article

Accurate above-ground biomass (AGB) quantification is confounded by signal saturation and data fusion challenges, particularly in structurally complex ecosystems like bamboo forests. To address these gaps, this study developed a two-stage framework to map the AGB of Dendrocalamus giganteus in a subtropical mountain environment. This study first employed Empirical Bayesian Kriging Regression Prediction (EBKRP) to spatialize sparse GEDI and ICESat-2 LiDAR metrics using Sentinel-2 and topographic covariates. Subsequently, a stacked ensemble model, integrating four machine learning algorithms, predicted AGB from the full suite of continuous variables. The stacking model achieved high predictive accuracy (R² = 0.84, RMSE = 11.07 Mg ha⁻¹) and substantially mitigated the common bias of underestimating high AGB, improving the predicted observed regression slope from a base model average of 0.63 to 0.81. Furthermore, SHAP analysis provided mechanistic insights, identifying the canopy photon rate as the dominant predictor and quantifying the ecological thresholds governing AGB distribution. The mean AGB density was 71.8 ± 21.9 Mg ha⁻¹, with its spatial pattern influenced by elevation and human settlements. This research provides a robust framework for synergizing multi-source remote sensing data to improve AGB estimation, offering a refined methodological pathway for large-scale carbon stock assessments. Full article

Monitoring the activity index of animals is crucial for assessing their welfare and behavior patterns. However, traditional methods for calculating the activity index, such as pixel intensity differencing of entire frames, are found to suffer from significant interference and noise, leading to inaccurate results. These classical approaches also do not support group or individual tracking in a user-friendly way, and no open-access platform exists for non-technical researchers. This study introduces an open-source web-based platform that allows researchers to calculate the activity index from top-view videos by selecting individual or group animals. It integrates Segment Anything Model2 (SAM2), a promptable deep learning segmentation model, to track animals without additional training or annotation. The platform accurately tracked Cobb 500 male broilers from weeks 1 to 7 with a 100% success rate, IoU of 92.21% ± 0.012, precision of 93.87% ± 0.019, recall of 98.15% ± 0.011, and F1 score of 95.94% ± 0.006, based on 1157 chickens. Statistical analysis showed that tracking 80% of birds in week 1, 60% in week 4, and 40% in week 7 was sufficient (r ≥ 0.90; p ≤ 0.048) to represent the group activity in respective ages. This platform offers a practical, accessible solution for activity tracking, supporting animal behavior analytics with minimal effort. Full article

The rise in online learning, accelerated by the COVID-19 pandemic, has led to greater use of synchronous hybrid learning (SHL) in higher education. SHL allows simultaneous teaching of in-person and online learners through videoconferencing tools. Previous studies have identified various benefits (e.g., flexibility) and challenges (e.g., student engagement) to SHL. Whilst systematic reviews have emerged on this topic, few studies have considered the experiences of staff. The aim of this review was threefold: (i) to better understand how staff experiences and perceptions are assessed, (ii) to understand staff experiences in terms of the benefits and challenges of SHL and (iii) to identify recommendations for effective teaching and learning using SHL. In line with the PRISMA guidance, we conducted a systematic review across four databases, identifying 14 studies for inclusion. Studies were conducted in nine different countries and covered a range of academic disciplines. Most studies adopted qualitative methods, with small sample sizes. Measures used were typically novel and unvalidated. Four themes were identified relating to (i) technology, (ii) redesigning teaching and learning, (iii) student engagement and (iv) staff workload. In terms of recommendations, ensuring adequate staff training and ongoing classroom support were considered essential. Additionally, active and collaborative learning were considered important to address issues with interactivity. Whilst these findings largely aligned with previous work, this review also identified limited reporting in research in this area, and future studies are needed to address this. Full article

Background: Accurate and timely prediction of mortality in intensive care unit (ICU) patients, particularly those with COVID-19, remains clinically challenging due to complex immune responses. Proteomic cytokine profiling holds promise for refining mortality risk assessment. Methods: Serum samples from 89 ICU patients (55 discharged, 34 deceased) were analyzed using a multiplex 21-cytokine panel. Samples were stratified into three groups based on time from collection to outcome: ≤48 h (Group 1: Early), >48 h to ≤7 days (Group 2: Intermediate), and >7 days to ≤14 days (Group 3: Late). Cytokine levels, simple cytokine ratios, and previously unexplored complex ratios between pro- and anti-inflammatory cytokines were evaluated. Machine learning-based feature selection identified the most predictive ratios, with performance evaluated by area under the curve (AUC), sensitivity, and specificity. Results: Complex cytokine ratios demonstrated superior predictive accuracy compared to traditional severity markers (APACHE II, SAPS II, SOFA), individual cytokines, and simple ratios, effectively distinguishing discharged from deceased patients across all groups (AUC: 0.918–1.000; sensitivity: 0.826–1.000; specificity: 0.775–0.900). Conclusions: Multiplex cytokine profiling enhanced by computationally derived complex ratios may offer robust predictive capabilities for ICU mortality risk stratification, serving as a valuable tool for personalized prognosis in critical care. Full article

Monitoring sediment concentration in water bodies is crucial for assessing water quality, ecosystems, and environmental health. However, physical sampling and sensor-based approaches are labor-intensive and unsuitable for large-scale, continuous monitoring. This study employs machine learning models to estimate suspended sediment concentration using images captured in natural light, named RGB, and near-infrared (NIR) conditions. A controlled dataset of approximately 1300 images with SSC values ranging from 1000 mg/L to 150,000 mg/L was developed, incorporating temperature, time of image capture, and solar irradiance as additional features. Random forest regression and gradient boosting regression were trained on mean RGB values, red reflectance, time of captured, and temperature for natural light images, achieving up to 72.96% accuracy within a 30% relative error. In contrast, NIR images leveraged gray-level co-occurrence matrix texture features and temperature, reaching 83.08% accuracy. Comparative analysis showed that ensemble models outperformed deep learning models like Convolutional Neural Networks and Multi-Layer Perceptrons, which struggled with high-dimensional feature extraction. These findings suggest that using machine learning models and RGB and NIR imagery offers a scalable, non-invasive, and cost-effective way of sediment monitoring in support of water quality assessment and environmental management. Full article

Adversarial attacks in Natural Language Processing (NLP) present a critical challenge, particularly in sentiment analysis, where subtle input modifications can significantly alter model predictions. In search of more robust defenses against adversarial attacks on sentimental analysis, this research work introduces two novel defense mechanisms: the Lexicon-Based Random Substitute Model (LRSM) and the Word-Variant Voting Model (WVVM). LRSM employs randomized substitutions from a dataset-specific lexicon to generate diverse input variations, disrupting adversarial strategies by introducing unpredictability. Unlike traditional defenses requiring synonym dictionaries or precomputed semantic relationships, LRSM directly substitutes words with random lexicon alternatives, reducing overhead while maintaining robustness. Notably, LRSM not only neutralizes adversarial perturbations but occasionally surpasses the original accuracy by correcting inherent model misclassifications. Building on LRSM, WVVM integrates LRSM, Frequency-Guided Word Substitution (FGWS), and Synonym Random Substitution and Voting (RS&V) in an ensemble framework that adaptively combines their outputs. Logistic Regression (LR) emerged as the optimal ensemble configuration, leveraging its regularization parameters to balance the contributions of individual defenses. WVVM consistently outperformed standalone defenses, demonstrating superior restored accuracy and F1 scores across adversarial scenarios. The proposed defenses were evaluated on two well-known sentiment analysis benchmarks: the IMDB Sentiment Dataset and the Yelp Polarity Dataset. The IMDB dataset, comprising 50,000 labeled movie reviews, and the Yelp Polarity dataset, containing labeled business reviews, provided diverse linguistic challenges for assessing adversarial robustness. Both datasets were tested using 4000 adversarial examples generated by established attacks, including Probability Weighted Word Saliency, TextFooler, and BERT-based Adversarial Examples. WVVM and LRSM demonstrated superior performance in restoring accuracy and F1 scores across both datasets, with WVVM excelling through its ensemble learning framework. LRSM improved restored accuracy from 75.66% to 83.7% when compared to the second-best individual model, RS&V, while the Support Vector Classifier WVVM variation further improved restored accuracy to 93.17%. Logistic Regression WVVM achieved an F1 score of 86.26% compared to 76.80% for RS&V. These findings establish LRSM and WVVM as robust frameworks for defending against adversarial text attacks in sentiment analysis. Full article

Research highlights a growing demand for active, experiential learning methods in higher education, especially in teacher education. While the benefits of parental involvement (PI) are well-documented, Hungary lacks tools to effectively prepare teacher trainees for fostering family–school cooperation. This study addresses this gap by introducing a custom-designed board game as an innovative teaching tool. The game simulates real-world challenges in PI through a cooperative, scenario-based framework. Exercises are grounded in international and national research, ensuring their relevance and evidence-based design. Tested with 110 students, the game’s educational value was assessed via post-gameplay questionnaires. Participants emphasized the strengths of its cooperative structure, realistic scenarios, and integration of humor. Many reported gaining new insights into parental roles and strategies for effective home–school partnerships. Practical applications include integrating the game into teacher education curricula and adapting it for other educational contexts. This study demonstrates how board games can bridge theory and practice, offering an engaging, effective medium to prepare future teachers for the challenges of PI. Full article

Landslides are destructive geological hazards, making accurate landslide susceptibility assessment essential for disaster prevention and mitigation. However, existing studies often lack scientific rigor in negative sample construction and have unclear model applicability. This study focuses on Ya’an City, Sichuan Province, China, and proposes an innovative approach to negative sample construction using Geographically Weighted Regression (GWR), which is then integrated with Tabular Network (TabNet), a deep learning architecture tailored to structured tabular data, to assess landslide susceptibility. The performance of TabNet is compared against Random Forest, Light Gradient Boosting Machine, deep neural networks, and Residual Networks. The experimental results indicate that (1) the GWR-based sampling strategy substantially improves model performance across all tested models; (2) TabNet trained using the GWR-based negative samples achieves superior performance over all other evaluated models, with an average AUC of 0.9828, exhibiting both high accuracy and interpretability; and (3) elevation, land cover, and annual Normalized Difference Vegetation Index are identified as dominant predictors through TabNet’s feature importance analysis. The results demonstrate that combining GWR and TabNet substantially enhances landslide susceptibility modeling by improving both accuracy and interpretability, establishing a more scientifically grounded approach to negative sample construction, and providing an interpretable, high-performing modeling framework for geological hazard risk assessment. Full article

Background: A detailed understanding of cardiac anatomy and physiology is crucial in cardiovascular medicine. However, traditional learning methods often fall short in addressing this complexity. Augmented reality (AR) offers a promising tool to enhance comprehension. To assess its potential integration into the Heidelberger Curriculum Medicinale (HeiCuMed), we conducted a needs assessment among medical students and lecturers at Heidelberg University Medical School. Methods: Our survey aimed to evaluate the perceived benefits of AR-based learning compared to conventional methods and to gather expectations regarding an AR course in cardiovascular medicine. Using LimeSurvey, we developed a questionnaire to assess participants’ prior AR experience, preferred learning methods, and interest in a proposed AR-based, 2 × 90-min in-person course. Results: A total of 101 students and 27 lecturers participated. Support for AR in small-group teaching was strong: 96.3% of students and 90.9% of lecturers saw value in a dedicated AR course. Both groups favored its application in anatomy, cardiac surgery, and internal medicine. Students prioritized congenital heart defects, coronary anomalies, and arrhythmias, while lecturers also emphasized invasive valve interventions. Conclusions: There is significant interest in AR-based teaching in cardiovascular education, suggesting its potential to complement and improve traditional methods in medical curricula. Further studies are needed to assess the potential benefits regarding learning outcomes. Full article

Stress in pigs poses significant challenges to animal welfare and productivity in modern pig farming, contributing to increased antimicrobial use and the rise of antimicrobial resistance (AMR). This study involves stress classification in pregnant sows by exploring five deep learning models: ConvNeXt, EfficientNet_V2, MobileNet_V3, RegNet, and Vision Transformer (ViT). These models are used for stress detection from facial images, leveraging an expanded dataset. A facial image dataset of sows was collected at Scotland’s Rural College (SRUC) and the images were categorized into primiparous Low-Stressed (LS) and High-Stress (HS) groups based on expert behavioural assessments and cortisol level analysis. The selected deep learning models were then trained on this enriched dataset and their performance was evaluated using cross-validation on unseen data. The Vision Transformer (ViT) model outperformed the others across the dataset of annotated facial images, achieving an average accuracy of 0.75, an F1 score of 0.78 for high-stress detection, and consistent batch-level performance (up to 0.88 F1 score). These findings highlight the efficacy of transformer-based models for automated stress detection in sows, supporting early intervention strategies to enhance welfare, optimize productivity, and mitigate AMR risks in livestock production. Full article

Background: Pancreatic cystic lesions (PCLs), including intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms (MCNs), pose a diagnostic challenge due to their variable malignant potential. Current guidelines, such as Fukuoka and American Gastroenterological Association (AGA), have moderate predictive accuracy and may lead to overtreatment or missed malignancies. Artificial intelligence (AI), incorporating machine learning (ML) and deep learning (DL), offers the potential to improve risk stratification, diagnosis, and management of PCLs by integrating clinical, radiological, and molecular data. This is the first systematic review to evaluate the application, performance, and clinical utility of AI models in the diagnosis, classification, prognosis, and management of pancreatic cysts. Methods: A systematic review was conducted in accordance with PRISMA guidelines and registered on PROSPERO (CRD420251008593). Databases searched included PubMed, EMBASE, Scopus, and Cochrane Library up to March 2025. The inclusion criteria encompassed original studies employing AI, ML, or DL in human subjects with pancreatic cysts, evaluating diagnostic, classification, or prognostic outcomes. Data were extracted on the study design, imaging modality, model type, sample size, performance metrics (accuracy, sensitivity, specificity, and area under the curve (AUC)), and validation methods. Study quality and bias were assessed using the PROBAST and adherence to TRIPOD reporting guidelines. Results: From 847 records, 31 studies met the inclusion criteria. Most were retrospective observational (n = 27, 87%) and focused on preoperative diagnostic applications (n = 30, 97%), with only one addressing prognosis. Imaging modalities included Computed Tomography (CT) (48%), endoscopic ultrasound (EUS) (26%), and Magnetic Resonance Imaging (MRI) (9.7%). Neural networks, particularly convolutional neural networks (CNNs), were the most common AI models (n = 16), followed by logistic regression (n = 4) and support vector machines (n = 3). The median reported AUC across studies was 0.912, with 55% of models achieving AUC ≥ 0.80. The models outperformed clinicians or existing guidelines in 11 studies. IPMN stratification and subtype classification were common focuses, with CNN-based EUS models achieving accuracies of up to 99.6%. Only 10 studies (32%) performed external validation. The risk of bias was high in 93.5% of studies, and TRIPOD adherence averaged 48%. Conclusions: AI demonstrates strong potential in improving the diagnosis and risk stratification of pancreatic cysts, with several models outperforming current clinical guidelines and human readers. However, widespread clinical adoption is hindered by high risk of bias, lack of external validation, and limited interpretability of complex models. Future work should prioritise multicentre prospective studies, standardised model reporting, and development of interpretable, externally validated tools to support clinical integration. Full article

To address the inefficiency of traditional concrete crack detection methods and the heavy reliance of supervised learning on extensive labeled data, in this study, an intelligent assessment method of concrete damage based on pseudo-label semi-supervised learning and fractal geometry theory is proposed to solve two core tasks: one is binary classification of pixel-level cracks, and the other is multi-category assessment of damage state based on crack morphology. Using three-channel RGB images as input, a dual-path collaborative training framework based on U-Net encoder–decoder architecture is constructed, and a binary segmentation mask of the same size is output to achieve the accurate segmentation of cracks at the pixel level. By constructing a dual-path collaborative training framework and employing a dynamic pseudo-label refinement mechanism, the model achieves an F1-score of 0.883 using only 50% labeled data—a mere 1.3% decrease compared to the fully supervised benchmark DeepCrack (F1 = 0.896)—while reducing manual annotation costs by over 60%. Furthermore, a quantitative correlation model between crack fractal characteristics and structural damage severity is established by combining a U-Net segmentation network with the differential box-counting algorithm. The experimental results demonstrate that under a cyclic loading of 147.6–221.4 kN, the fractal dimension monotonically increases from 1.073 (moderate damage) to 1.189 (failure), with 100% accuracy in damage state identification, closely aligning with the degradation trend of macroscopic mechanical properties. In complex crack scenarios, the model attains a recall rate (Re = 0.882), surpassing U-Net by 13.9%, with significantly enhanced edge reconstruction precision. Compared with the mainstream models, this method effectively alleviates the problem of data annotation dependence through a semi-supervised strategy while maintaining high accuracy. It provides an efficient structural health monitoring solution for engineering practice, which is of great value to promote the application of intelligent detection technology in infrastructure operation and maintenance. Full article

To improve collision risk prediction in high-traffic coastal waters and support real-time decision-making in maritime navigation, this study proposes a regional collision risk prediction system integrating the Computed Distance at Collision (CDC) method with an Adaptive Neuro-Fuzzy Inference System (ANFIS). Unlike Distance at Closest Point of Approach (DCPA), which depends on the position of Global Positioning System (GPS) antennas, Computed Distance at Collision (CDC) directly reflects the actual hull shape and potential collision point. This enables a more realistic assessment of collision risk by accounting for the hull geometry and boundary conditions specific to different ship types. The system was designed and validated using ship motion simulations involving bulk and container ships across varying speeds and crossing angles. The CDC method was used to define collision, almost-collision, and near-collision situations based on geometric and hydrodynamic criteria. Subsequently, the FIS–CDC model was constructed using the ANFIS by learning patterns in collision time and distance under each condition. A total of four input variables—ship speed, crossing angle, remaining time, and remaining distance—were used to infer the collision risk index (CRI), allowing for a more nuanced and vessel-specific assessment than traditional CPA-based indicators. Simulation results show that the time to collision decreases with higher speeds and increases with wider crossing angles. The bulk carrier exhibited a wider collision-prone angle range and a greater sensitivity to speed changes than the container ship, highlighting differences in maneuverability and risk response. The proposed system demonstrated real-time applicability and accurate risk differentiation across scenarios. This research contributes to enhancing situational awareness and proactive risk mitigation in Maritime Autonomous Surface Ship (MASS) and Vessel Traffic System (VTS) environments. Future work will focus on real-time CDC optimization and extending the model to accommodate diverse ship types and encounter geometries. Full article