Search Results (1,792)

This study explores fine-tuning methods and dataset structures for multilingual neural machine translation using the No Language Left Behind model, with a case study on Kazakh, English, and Russian. We compare single-stage and two-stage fine-tuning approaches, as well as triplet versus non-triplet dataset configurations, to improve translation quality. A high-quality, 50,000-triplet dataset in information technology domain, manually translated and expert-validated, serves as the in-domain benchmark, complemented by out-of-domain corpora like KazParC. Evaluations using BLEU, chrF, METEOR, and TER metrics reveal that single-stage fine-tuning excels for low-resource pairs (e.g., 0.48 BLEU, 0.77 chrF for Kazakh → Russian), while two-stage fine-tuning benefits high-resource pairs (Russian → English). Triplet datasets improve cross-linguistic consistency compared with non-triplet structures. Our reproducible framework offers practical guidance for adapting neural machine translation to technical domains and low-resource languages. Full article

Effective energy management in Low Earth Orbit satellites is critical, as inefficient energy management can significantly affect mission objectives. The dynamic and harsh space environment further complicates the development of effective energy management strategies. To address these challenges, we propose a Deep Reinforcement Learning approach using Deep-Q Network to develop an adaptive energy management framework for Low Earth Orbit satellites. Compared to traditional techniques, the proposed solution autonomously learns from environmental interaction, offering robustness to uncertainty and online adaptability. It adjusts to changing conditions without manual retraining, making it well-suited for handling modeling uncertainties and non-stationary dynamics typical of space operations. Training is conducted using a realistic satellite electric power system model with accurate component parameters and single-orbit power profiles derived from real space missions. Numerical simulations validate the controller performance across diverse scenarios, including multi-orbit settings, demonstrating superior adaptability and efficiency compared to conventional Maximum Power Point Tracking methods. Full article

Adverse drug events (ADEs) significantly impact patient safety and health outcomes. Manual ADE detection from clinical narratives is time-consuming, labor-intensive, and costly. Recent advancements in large language models (LLMs), including transformer-based architectures such as Bidirectional Encoder Representations from Transformers (BERT) and Generative Pretrained Transformer (GPT) series, offer promising methods for automating ADE extraction from clinical data. These models have been applied to various aspects of pharmacovigilance and clinical decision support, demonstrating potential in extracting ADE-related information from real-world clinical data. Additionally, chatbot-assisted systems have been explored as tools in clinical management, aiding in medication adherence, patient engagement, and symptom monitoring. This narrative review synthesizes the current state of LLMs in ADE detection from a clinical perspective, organizing studies into categories such as human-facing decision support tools, immune-related ADE detection, cancer-related and non-cancer-related ADE surveillance, and personalized decision support systems. In total, 39 articles were included in this review. Across domains, LLM-driven methods have demonstrated promising performances, often outperforming traditional approaches. However, critical limitations persist, such as domain-specific variability in model performance, interpretability challenges, data quality and privacy concerns, and infrastructure requirements. By addressing these challenges, LLM-based ADE detection could enhance pharmacovigilance practices, improve patient safety outcomes, and optimize clinical workflows. Full article

Rockfill particle gradation significantly influences mechanical performance in earth–rockfill dam construction, yet on-site screening is often time-consuming, labor-intensive, and structurally invasive. This study proposes a rapid and non-destructive detection method using mobile-based photography and an end-to-end image segmentation approach. An enhanced YOLOv8-seg model with an integrated dual-attention mechanism was pre-trained on laboratory images to accurately segment densely stacked particles. Transfer learning was then employed to retrain the model using a limited number of on-site images, achieving high segmentation accuracy. The proposed model attains a mAP50 of 97.8% (base dataset) and 96.1% (on-site dataset), enabling precise segmentation of adhered and overlapped particles with various sizes. A Minimum Area Rectangle algorithm was introduced to compute the gradation, closely matching the results from manual screening. This method significantly contributes to the automation of construction workflows, cutting labor costs, minimizing structural disruption, and ensuring reliable measurement quality in earth–rockfill dam projects. Full article

Background: Autism spectrum disorder (ASD) represents a neurobiological disorder with a high prevalence in the children’s population. The aim of the present review was to assess the current evidence on the use of salivary biomarkers for the early diagnosis of ASD. Materials and methods: A search was conducted on the electronic databases PUBMED/Medline, Google Scholar and Scopus for the retrieval of articles concerning the study topic. Results: A total of 22 studies have been included in the present review considering 21 articles identified from databases and 1 article included using a manual search. A wide range of biomarkers have been proposed for early detection of ASD diseases including nonspecific inflammation markers like interleukin-1β (IL-1β), interleukin-6 (IL-6), interleukin-8 (IL-8), tumor necrosis factor α (TNFα), oxidative stress markers like superoxide dismutase and glutathione peroxidase, hormones such as cortisol and oxytocin, various microRNAs including miR-21, miR-132 and miR-137, and exosomes. The techniques used for biomarke detection may vary according to molecule type and concentration. Conclusions: salivary biomarkers could represent a potential useful tool for the primary detection of several systemic diseases including ASD, taking advantage of non-invasiveness and cost-effective capability compared to other biofluid-based diagnostic techniques. Full article

Amylose content significantly influences the technological, quality, and nutritional properties of starchy foods. This study developed a rapid, non-destructive method to quantify amylose content in starch using mid-infrared (MIR) spectroscopy combined with chemometric techniques. Manually prepared starch mixtures with varying amylose levels were scanned to obtain MIR spectra, which were preprocessed using smoothing and z-score normalization to reduce operational variability. Three variable selection methods, including bootstrap soft shrinkage (BOSS), competitive adaptive reweighted sampling (CARS), and uninformative variable elimination (UVE), were applied to select the useful spectra. A partial least square (PLS) model was then constructed to correlate selected spectral data with amylose content. The results revealed that the number and position of selected variables differed across different optimization methods, which influenced the model’s performance. It is worth noting that the optimized PLS model significantly reduced the root mean squared error of cross-validation (RMSECV) and improved prediction accuracy in 50 runs. In particular, the CARS-PLS model showed superior performance, achieving a correlation coefficient (R_p) of 0.964 and a root mean squared error of prediction (RMSEP) of 4.59, a 60% improvement over the original PLS model, which had an RMSEP of 11.56. These results highlight MIR spectroscopy’s potential, combined with optimized chemometric models, for accurate amylose quantification in food quality control. Full article

Monitoring position and movements of preterm infants is important to ensure their well-being and optimal development. This study evaluates the feasibility of a pressure-sensitive fiber-optic mattress (FM), made entirely of plastic, for two-dimensional analysis of preterm infant movements and positioning. Before clinical use, we developed a simple, replicable, and cost-effective test protocol to simulate infant movements and positions, enabling early identification of technical limitations. Using data from 20 preterm infants, we assessed the FM’s potential to monitor posture and limb motion. FM-derived pressure patterns were compared with camera-based manual annotations to distinguish between different positions and out-of-bed moments, as well as limb-specific movements. Bench-test results demonstrated the FM’s sensitivity to motion and pressure changes, supporting its use in preclinical validation. Clinical data confirmed the FM’s reliability in identifying infant positions and movement patterns, showing an accuracy comparable to camera annotations. However, limitations such as calibration, sensitivity to ambient light, and edge-related artifacts were noted, indicating areas for improvement. In conclusion, the test protocol proved effective for early-stage evaluation of smart mattress technologies. The FM showed promising clinical feasibility for non-obtrusive monitoring of preterm infants, though further optimization is needed for robust performance in neonatal care. Full article

Objective: We evaluated the feasibility of a machine-learning (ML) model based on clinical features and radiomics from [¹⁸F]FDG PET/CT images to differentiate between infected and non-infected intracavitary vascular grafts and endografts (iVGEI). Methods: Three ML models were developed: one based on pre-treatment criteria to diagnose a vascular graft infection (“MAGIC-light features”), another using radiomics features from diagnostic [¹⁸F]FDG-PET scans, and a third combining both datasets. The training set included 92 patients (72 iVGEI-positive, 20 iVGEI-negative), and the external test set included 20 iVGEI-positive and 12 iVGEI-negative patients. The abdominal aorta and iliac arteries in the PET/CT scans were automatically segmented using SEQUOIA and TotalSegmentator and manually adjusted, extracting 96 radiomics features. The best-performing models for the MAGIC-light features and PET-radiomics features were selected from 343 unique models. Most relevant features were combined to test three final models using ROC analysis, accuracy, sensitivity, and specificity. Results: The combined model achieved the highest AUC in the test set (mean ± SD: 0.91 ± 0.02) compared with the MAGIC-light-only model (0.85 ± 0.06) and the PET-radiomics model (0.73 ± 0.03). The combined model also achieved a higher accuracy (0.91 vs. 0.82) than the diagnosis based on all the MAGIC criteria and a comparable sensitivity and specificity (0.70 and 1.00 vs. 0.76 and 0.92, respectively) while providing diagnostic information at the initial presentation. The AUC for the combined model was significantly higher than the PET-radiomics model (p = 0.02 in the bootstrap test), while other comparisons were not statistically significant. Conclusions: This study demonstrated the potential of ML models in supporting diagnostic decision making for iVGEI. A combined model using pre-treatment clinical features and PET-radiomics features showed high diagnostic performance and specificity, potentially reducing overtreatment and enhancing patient outcomes. Full article

Background and Objectives: The management of type 2 diabetes (T2D) extends beyond glycemic control, requiring a more global strategy that includes optimization of body composition, even more so in the context of sarcopenia and visceral adiposity, as they contribute to poor outcomes. Past reviews have typically been focused on weight reduction or glycemic effectiveness, with limited inclusion of new therapies’ effects on muscle and fat distribution. In addition, the emergence of incretin-based therapies and dual agonists such as tirzepatide requires an updated synthesis of their impacts on body composition. This review attempts to bridge the gap by taking a systematic approach to how current blood-glucose lowering therapies affect lean body mass, fat mass, and the risk of sarcopenia in T2D patients. Materials and Methods: Between January 2015 and March 2025, we conducted a narrative review by searching the PubMed, Scopus, and Web of Science databases for English-language articles. The keywords were combinations of the following: “type 2 diabetes,” “lean body mass,” “fat mass,” “body composition,” “sarcopenia,” “GLP-1 receptor agonists,” “SGLT2 inhibitors,” “tirzepatide,” and “antidiabetic pharmacotherapy.” Reference lists were searched manually as well. The highest precedence was assigned to studies that aimed at adult type 2 diabetic subjects and reported body composition results. Inclusion criteria for studies were: (1) type 2 diabetic mellitus adult patients and (2) reporting measures of body composition (e.g., lean body mass, fat mass, or muscle function). We prioritized randomized controlled trials and large observational studies and excluded mixed diabetic populations, non-pharmacological interventions only, and poor reporting of body composition. Results: Metformin was widely found to be weight-neutral with minimal effects on muscle mass. Insulin therapy, being an anabolic hormone, often leads to fat mass accumulation and increases the risk of sarcopenic obesity. Incretin-based therapies induced substantial weight loss, mostly from fat mass. Notable results were observed in studies with tirzepatide, demonstrating superior reduction not only in fat mass, but also in visceral fat. Sodium-glucose cotransporter 2 inhibitors (SGLT2 inhibitors) promote fat loss but are associated with a small yet significant decrease in lean muscle mass. Conclusions: Blood-glucose lowering therapies demonstrated clinically relevant effects on body composition. Treatment should be personalized, balancing glycemic control, cardiovascular, and renal benefits, together with optimal impact on muscle mass along with glycemic, cardiovascular, and renal benefits. Full article

Early printed books, particularly incunabula, are invaluable archives of the beginnings of modern educational systems. However, their complex layouts, antique typefaces, and page degradation caused by bleed-through and ink fading pose significant challenges for automatic transcription. In this work, we present a modular pipeline that addresses these problems by combining modern layout analysis and language modeling techniques. The pipeline begins with historical layout-aware text segmentation using Kraken, a neural network-based tool tailored for early typographic structures. Initial optical character recognition (OCR) is then performed with Kraken’s recognition engine, followed by post-correction using a fine-tuned ByT5 transformer model trained on manually aligned line-level data. By learning to map noisy OCR outputs to verified transcriptions, the model substantially improves recognition quality. The pipeline also integrates a preprocessing stage based on our previous work on bleed-through removal using robust statistical filters, including non-local means, Gaussian mixtures, biweight estimation, and Gaussian blur. This step enhances the legibility of degraded pages prior to OCR. The entire solution is open, modular, and scalable, supporting long-term preservation and improved accessibility of cultural heritage materials. Experimental results on 15th-century incunabula show a reduction in the Character Error Rate (CER) from around 38% to around 15% and an increase in the Bilingual Evaluation Understudy (BLEU) score from 22 to 44, confirming the effectiveness of our approach. This work demonstrates the potential of integrating transformer-based correction with layout-aware segmentation to enhance OCR accuracy in digital humanities applications. Full article

Monitoring crop damage caused by wildlife remains a significant challenge in agricultural management, particularly in the case of large-scale monocultures such as maize. The given study presents a semi-automated process for detecting wildlife-induced damage using RGB imagery acquired from unmanned aerial vehicles (UAVs). The method is designed for non-specialist users and is fully integrated within the QGIS platform. The proposed approach involves calculating three vegetation indices—Excess Green (ExG), Green Leaf Index (GLI), and Modified Green-Red Vegetation Index (MGRVI)—based on a standardized orthomosaic generated from RGB images collected via UAV. Subsequently, an unsupervised k-means clustering algorithm was applied to divide the field into five vegetation vigor classes. Within each class, 25% of the pixels with the lowest average index values were preliminarily classified as damaged. A dedicated QGIS plugin enables drone data analysts (Drone Data Analysts—DDAs) to adjust index thresholds, based on visual interpretation, interactively. The method was validated on a 50-hectare maize field, where 7 hectares of damage (15% of the area) were identified. The results indicate a high level of agreement between the automated and manual classifications, with an overall accuracy of 81%. The highest concentration of damage occurred in the “moderate” and “low” vigor zones. Final products included vigor classification maps, binary damage masks, and summary reports in HTML and DOCX formats with visualizations and statistical data. The results confirm the effectiveness and scalability of the proposed RGB-based procedure for crop damage assessment. The method offers a repeatable, cost-effective, and field-operable alternative to multispectral or AI-based approaches, making it suitable for integration with precision agriculture practices and wildlife population management. Full article

The increasing utilization of hard-to-cut materials in high-performance sectors such as aerospace and defense has pushed manufacturing systems to be flexible in processing large workpieces with a wide range of materials while also delivering high precision. Recent studies have highlighted the potential of integrating industrial robots (IRs) with electric discharge machining (EDM) to create a non-contact, low-force manufacturing platform, particularly suited for the accurate machining of hard-to-cut materials into complex and large-scale monolithic components. In response to this potential, a novel robotic EDM system has been developed. However, the manual programming and control of such a convoluted system present a significant challenge, often leading to inefficiencies and increased error rates, creating a scenario where the EDM process becomes unfeasible. To enhance the industrial applicability of this robotic EDM technology, this study focuses on a novel methodology to develop and validate a digital twin (DT) of the physical robotic EDM system. The digital twin functions as a virtual experimental environment for tool motion, effectively addressing the challenges posed by collisions and kinematic singularities inherent in the physical system, yet with proven 20-micron EDM gap accuracy. Furthermore, it facilitates a CNC-like, user-friendly offline programming framework for robotic EDM cutting path generation. Full article

Inter-turn short-circuit (ITSC) faults in permanent magnet synchronous machines (PMSMs) present a significant reliability challenge in electric vehicle (EV) drivetrains, particularly under non-stationary operating conditions characterized by inverter-driven transients, variable loads, and magnetic saturation. Existing diagnostic approaches, including motor current signature analysis (MCSA) and wavelet-based methods, are primarily designed for steady-state conditions and rely on manual feature selection, limiting their applicability in real-time embedded systems. Furthermore, the lack of publicly available, high-fidelity datasets capturing the transient dynamics and nonlinear flux-linkage behaviors of PMSMs under fault conditions poses an additional barrier to developing data-driven diagnostic solutions. To address these challenges, this study introduces a simulation framework that generates a comprehensive dataset using finite element method (FEM) models, incorporating magnetic saturation effects and inverter-driven transients across diverse EV operating scenarios. Time-frequency features extracted via Discrete Wavelet Transform (DWT) from stator current signals are used to train a Transformer model for automated ITSC fault detection. The Transformer model, leveraging self-attention mechanisms, captures both local transient patterns and long-range dependencies within the time-frequency feature space. This architecture operates without sequential processing, in contrast to recurrent models such as LSTM or RNN models, enabling efficient inference with a relatively low parameter count, which is advantageous for embedded applications. The proposed model achieves 97% validation accuracy on simulated data, demonstrating its potential for real-time PMSM fault detection. Additionally, the provided dataset and methodology contribute to the facilitation of reproducible research in ITSC diagnostics under realistic EV operating conditions. Full article

This study proposes a non-semantic multimodal approach to predict segment access frequency (SAF) in lecture archives. Such archives, widely used as supplementary resources in modern education, often consist of long, unedited recordings that are difficult to navigate and review efficiently. The predicted SAF, an indicator of student viewing behavior, serves as a practical proxy for student engagement. The increasing volume of recorded material renders manual editing and annotation impractical, making the automatic identification of high-SAF segments crucial for improving accessibility and supporting targeted content review. The approach focuses on lecture archives from a real-world blended learning context, characterized by resource constraints such as no specialized hardware and limited student numbers. The model integrates multimodal features from instructor’s actions (via OpenPose and optical flow), audio spectrograms, and slide page progression—a selection of features that makes the approach applicable regardless of lecture language. The model was evaluated on 665 labeled one-minute segments from one such course. Experiments show that the best-performing model achieves a Pearson correlation of 0.5143 in 7-fold cross-validation and 61.05% average accuracy in a downstream three-class classification task. These results demonstrate the system’s capacity to enhance lecture archives by automatically identifying key segments, which aids students in efficient, targeted review and provides instructors with valuable data for pedagogical feedback. Full article

Objectives: This pilot study evaluates the correlation between periodontal pocket depth (PPD) measurements obtained by manual probing and those derived from an AI-coupled ultrasound imaging device in periodontitis patients. Materials and Methods: Thirteen patients with periodontitis underwent ultrasonic probing with an AI engine for automated PPD measurements, followed by routine manual probing. Results: A total of 2088 manual and 1987 AI-based PPD measurements were collected. The mean PPD was 4.2 mm (range: 2–8 mm) for manual probing and 4.5 mm (range: 2–9 mm) for AI-based ultrasound, with a Pearson correlation coefficient of 0.68 (95% CI: 0.62–0.73). Discrepancies were noted in cases with inflammation or calculus. AI struggled to differentiate pocket depths in complex clinical scenarios. Discussion: Ultrasound imaging offers non-invasive, real-time visualization of periodontal structures, but AI accuracy requires further training to address image artifacts and clinical variability. Conclusions: The ultrasound device shows promise for non-invasive periodontal diagnostics but is not yet a direct alternative to manual probing. Further AI optimization and validation are needed. Clinical Relevance: This technology could enhance patient comfort and enable frequent monitoring, pending improvements in AI reliability. Full article