Search Results (4,548)

Background: Upper tract urothelial carcinoma (UTUC) is a rare malignancy representing approximately 5–10% of all urothelial cancers. Key risk factors include smoking, chemical exposures, selected metabolic conditions, and hereditary cancer syndromes. This narrative review summarises current knowledge on UTUC molecular pathogenesis, major risk determinants, and contemporary therapeutic strategies. Methods: This study was conducted as a narrative review with a structured literature search. PubMed, Web of Science, Embase, and Scopus were searched using predefined combinations of UTUC-related terms covering molecular pathogenesis, carcinogenic risk factors, and treatment strategies. The review was prepared according to SANRA principles to improve transparency and consistency; however, no formal systematic review methodology or meta-analysis was performed. Results: Available genomic studies indicate that UTUC has a molecular profile distinct from urothelial bladder carcinoma (UBC), with recurrent alterations involving FGFR3, HRAS, KMT2D, CDKN2A, KRAS, MYC, and BRIP1. Smoking, aristolochic acid exposure, Lynch syndrome, and possibly early-onset urolithiasis contribute to carcinogenesis through distinct but incompletely understood mechanisms. Surgical treatment remains the standard of care for high-risk localised disease, whereas perioperative chemotherapy, immunotherapy, and targeted agents are expanding treatment options, particularly in advanced disease. A substantial proportion of the therapeutic evidence, however, is derived from broader urothelial carcinoma populations rather than UTUC-specific studies. Conclusions: UTUC is biologically heterogeneous and shaped by both molecular alterations and environmental exposures. Although substantial progress has been made, important gaps remain in understanding UTUC-specific carcinogenic mechanisms and in defining evidence-based personalised treatment strategies. Better integration of molecular, environmental, and clinical data is needed to improve risk stratification and treatment selection. Full article

Background and objective: During adolescence, physical fitness attributes are a key factor positively influencing not only physical health, but also cognitive, psychological and social development. The objective was to derive data-driven physical fitness profiles and to characterize their sex-specific structure in Chilean adolescents using unsupervised machine learning and an explainable artificial intelligence (AI) framework. Methods: This cross-sectional study included a nationally representative sample of 25,912 Chilean schoolchildren (14,184 boys and 11,728 girls) aged 13 to 17 years. Six unsupervised clustering algorithms were applied separately by sex, each configured to partition participants into three clusters representing low, medium, and high physical performance levels. Results: K-means achieved the highest composite internal validity score for both sexes (boys: 0.959, Calinski–Harabasz 5757.7, Davies–Bouldin 1.203; girls: 0.976, Calinski–Harabasz 4156.1, Davies–Bouldin 1.349), and the partition was highly stable across 500 age-stratified bootstrap resamples (adjusted Rand index 0.973 in boys and 0.967 in girls). A Random-Forest surrogate reproduced the cluster labels with high fidelity (macro-averaged F1 0.984 in both sexes; Cohen’s κ 0.973 in boys and 0.975 in girls). Conclusions: Unsupervised machine learning with explainable AI provides a robust, transparent, sex-specific framework for stratifying adolescent fitness; cardiorespiratory capacity and abdominal endurance are the components that most consistently distinguish the profiles, while lower-limb power and upper-body endurance differentiate the higher performance tiers. Full article

Congenital anomalies of the kidney and urinary tract (CAKUT) are the leading cause of pediatric chronic kidney disease (CKD) and comprise a heterogeneous group of developmental disorders with a substantial genetic contribution. Advances in next-generation sequencing have facilitated the identification of numerous candidate genes and rare variants associated with CAKUT. However, establishing causality and defining the biological functions of implicated genes remain major challenges. Functional validation is therefore essential to bridge the gap between gene discovery and mechanistic understanding, enabling the interpretation of genetic variation within the context of kidney development and disease. The zebrafish (Danio rerio) has emerged as a powerful in vivo model for studying renal development and interrogating the function of CAKUT-associated genes. Its utility stems from a high degree of genetic and developmental conservation with humans, conserved nephrogenic pathways, optical transparency during embryogenesis, and the relative ease of genetic manipulation. In this review, we provide an overview of zebrafish kidney development within the broader context of vertebrate nephrogenesis, highlighting the key genetic programs governing intermediate mesoderm specification, nephron segmentation, and pronephric morphogenesis. We then systematically examine CAKUT-associated genes that have been modeled in zebrafish, focusing on studies that have linked genetic perturbations to renal development and structural phenotypes. Finally, we discuss the strengths and limitations of zebrafish models for functional genomics and variant interpretation and consider their emerging role in bridging genetic discovery with mechanistic insights into CAKUT pathogenesis. Full article

This study proposes a sequential Hybrid GP-ANN architecture based on the concept of autonomous feature engineering for multi-output performance prediction of single-phase induction motors, serving as a high-fidelity surrogate model ready for digital twin integration. The required high-resolution motor dataset (88,375 samples) was generated using a parametric sweep approach with Ansys RMxprt. In the proposed architecture, Genetic Programming (GP) is not positioned as a final predictor but as an analytical filter that discovers hidden physical relationships in raw input data and converts them into super features, thereby structurally eliminating the polynomial explosion risk and scale sensitivity that arise when raw data are modelled directly. The nonlinear physical relationships discovered autonomously by GP are added to the network input matrix as mathematical vectors, breaking the black-box structure of standard artificial neural networks and transforming it into a physics-inspired grey-box model. The nonlinear terms discovered by GP are fed into the network after independent Z-score normalisation to preserve gradient stability. To comprehensively evaluate this framework, its predictive performance is benchmarked against industry-standard machine learning algorithms, including Random Forest (RF), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost). Results on variation-based unseen test data show that the hybrid model achieves competitive prediction accuracy relative to XGBoost—matching or exceeding it for Output Torque and Input Current, and remaining broadly comparable for Output Power, though XGBoost achieves substantially lower RMSE for Efficiency and Power Factor—while additionally offering transparent mathematical traceability, reducing error rates (RMSE) by 42.1% to 66.1% per parameter compared to the standard ANN. The model achieves an $R^2$ score of 0.9879 for the power factor parameter, where conventional approaches struggle. To rigorously validate the interpretability of this grey-box architecture, a global Permutation Feature Importance (PFI) analysis was conducted, showing that the GP-derived super features collectively account for 61.46% of the decision logic, outweighing the combined contribution of the raw inputs (38.54%). Furthermore, the autonomous feature engineering layer is found to reduce the learning burden on the ANN, allowing it to converge to a substantially lower error floor within the same fixed epoch budget. With an online inference time below 0.02 ms, the proposed architecture offers a robust methodological infrastructure for sustainable motor digital twins through a balanced trade-off between prediction accuracy and computational efficiency. Full article

This paper presents an explainable defense framework against perception-layer and Man-in-the-Middle (MitM) attacks in Internet of Things (IoT)-based environmental hazard warning systems. These systems rely on heterogeneous sensors (gas, light, sound, temperature, and humidity) whose integrity is crucial for reliable environmental alerts. Perception-layer attacks such as spoofing, jamming, and data injection can compromise sensor readings, while MitM attacks threaten communication reliability. The proposed approach integrates incremental Dynamic Time Warping (DTW) for time-series anomaly detection with a tree- based ensemble classifier (XGBoost), in addition to Shapley Additive Explanations (SHAP) for interpretability. A comparative evaluation framework jointly considers detection performance and explanation quality through metrics including pre-registering a Casual Ground Truth based on network protocol localized Precision@ K feature overlap metrics (Q), instead of relying on subjective human-expert or global rank correlations to quantitively evaluate the explanation transparency. Experimental simulations using an authentic EdgeIIoT-2022 dataset under 3-fold forward–chaining cross-validation demonstrated high detection accuracy and moderated explainability scores. The results prove the framework’s ability to detect and explain adversarial behaviors in sensor networks, strengthening trust, transparency, and resilience in safety-critical IoT infrastructures. Full article

Background: Keratoconus is a progressive corneal ectasia characterised by extracellular matrix (ECM) loss and an emerging inflammatory component, for which no disease-modifying molecular therapy exists. Exosomes derived from limbal and mesenchymal stem cells are an attractive cell-free therapeutic modality, but the cargo that should be delivered is undefined, and no curated limbal stem-cell (LSC) exosome cargo dataset currently exists. Methods: We reanalysed a public keratoconus corneal RNA-sequencing dataset (GEO: GSE77938; discovery and replication cohorts) with DESeq2, defined a replicated differentially expressed gene (DEG) set, and performed Gene Ontology, KEGG and Reactome enrichment. A high-confidence protein–protein interaction (PPI) network (STRING) identified hub genes. We integrated keratoconus disease-gene evidence (Open Targets Platform) and documented extracellular-vesicle cargo (ExoCarta, Vesiclepedia) and computed a transparent Cargo Prioritization Score (CPS) to nominate candidate LSC-exosome therapeutic cargo. Results: A total of 1677 DEGs were detected in discovery (152 up, 1525 down) and 1380 were replicated. Enrichment was dominated by extracellular matrix organisation; adaptive immune response; and mononuclear cell differentiation. Network analysis nominated ECM and immune hub genes. The CPS prioritised COL1A1, FN1, COL4A1, COL3A1, COL5A1, MMP1 as leading restoration-cargo candidates, all documented as EV cargo and present in the mesenchymal stem-cell EV reference proteome. Conclusions: This fully reproducible, real-data framework provides a ranked, evidence-traceable shortlist of candidate LSC-exosome cargo for keratoconus and an explicit account of current data gaps to guide experimental validation. Full article

Cardiovascular disease (CVD) remains the leading cause of global mortality, underscoring the urgent need for predictive frameworks that are both accurate and clinically interpretable. This study introduces a hybrid diagnostic framework that integrates the novel Competitive Learning and Past-Based Optimization (CLPO) algorithm with the XGBoost classifier to enhance CVD risk prediction and biomarker identification. Unlike conventional optimization-based approaches, CLPO employs past-informed learning, competitive interactions, and an adaptive escape mechanism to dynamically balance exploration and exploitation, ensuring stable convergence and efficient hyperparameter tuning. Experimental evaluations on benchmark cardiovascular datasets achieved an impressive 94.79% cross-validation accuracy, F1-score of 0.9509, and AUC of 0.965, with a recall of 0.955, confirming the model’s ability to minimize false negatives—a critical factor in clinical screening. The framework also demonstrated computational efficiency, with an average runtime of 203.8 ± 57.4 s, ranking first overall among nine comparative optimizers. SHAP-based explainability analysis revealed ST slope, chest pain type, and cholesterol as the most predictive risk factors, offering transparent and clinically meaningful interpretations. CLPO-XGBoost consistently achieved highly competitive diagnostic accuracy and superior algorithmic stability compared with prominent recent state-of-the-art frameworks. These findings position the proposed framework as a next-generation predictive cardiology model that unites high diagnostic accuracy, computational efficiency, and interpretability—bridging the gap between artificial intelligence and clinical decision support. Full article

Driver behavior detection is essential for improving road safety in Intelligent Transportation Systems (ITSs). This paper proposes an attention-enhanced CNN integrated with Explainable Artificial Intelligence (XAI) to achieve reliable and interpretable driver behavior classification. Furthermore, the model also uses a multi-head attention mechanism to ensure that it captures local spatial features as well as long-range dependencies in the activities of the driver. The State Farm Distracted Driver dataset is experimented with using stratified 5-fold cross-validation. The proposed MHA-CNN model has a mean accuracy of 99.48% on all folds and an overall high levels of precision, recall, and F1-score in all ten distraction classes. Grad-CAM is used to produce attribution maps to improve interpretability by showing semantically important parts of the image, including hands, face, and mobile devices, and can give a visual explanation of why models make decisions. In addition, an ablation investigation is performed with k-fold validation to assess the value of the multi-head attention mechanism. Its model can run at 122 frames per second with 42.92 million parameters, showing that the suggested method is a reasonable tradeoff between accuracy, efficiency, and transparency and can be applied in real-time driver monitoring applications. Full article

In this paper, we report a novel self-smoothed gradient index cladding waveguide for low-loss, high-density photonic integrated circuits (PICs). In conventional PIC waveguides, there is a fundamental trade-off between propagation loss and bending loss. High-index-contrast waveguides can provide strong mode confinement and small bending radius, but they are highly sensitive to sidewall roughness and therefore exhibit increased scattering loss. In contrast, low-index-contrast waveguides can reduce propagation loss, but they require a much larger bending radius and are not suitable for dense photonic integration. To overcome this limitation, we propose a self-smoothed double-cladding waveguide architecture composed of a high-index core, a gradient index first cladding layer, and a low-index second cladding layer. The first cladding layer can be formed by advanced conformal coating processes, such as non-uniformly cycled atomic layer deposition and gradient index dip coating, which provide both a gradual refractive index transition and a self-smoothing effect on the rough sidewall. We perform quantitative analyses of the propagation loss, bending loss, and mode field distribution of the proposed structure. The numerical results confirm that the propagation loss can be reduced by approximately two orders of magnitude while maintaining low bending loss and mode confinement comparable to that of a conventional rib-shaped waveguide. We also experimentally verify the self-smoothing effect by using atomic layer deposition (ALD)-grown non-uniformly cycled nanolaminates and coating a rough sapphire bar with a high-refractive-index polymer, which significantly reduces the measured surface roughness and improves optical transparency. These results confirm that the proposed self-smoothed gradient index cladding waveguide can be an effective platform for realizing low-loss, high-density PICs. Full article

Despite growing interest in artificial intelligence (AI) applications in the construction industry, the literature still lacks a consolidated understanding of how AI functions across the full spectrum of construction supply chain processes. Existing studies are dispersed across different technologies, project stages, and application contexts, making it difficult to identify the intellectual structure of this field, the main areas of AI application, and the barriers that continue to constrain practical implementation. To address this gap, this study conducts a systematic review of AI applications in construction supply chains by combining scientometric analysis with qualitative content synthesis. A total of 212 journal articles retrieved from Scopus were analyzed using VOSviewer-based scientometric analysis and qualitative content synthesis. The scientometric analysis maps annual publication trends, keyword co-occurrence patterns, co-cited sources, influential documents, and collaboration networks. The qualitative synthesis further examines how AI supports construction supply chain management across three broad themes: procurement and production optimization, logistics and material management, and collaborative decision-making for resilience and sustainability. The findings show that AI has been primarily applied to demand forecasting, resource optimization, logistics coordination, contract and document processing, computer vision-based monitoring, and multi-agent decision support. However, its practical diffusion remains constrained by fragmented and low-quality data, limited empirical validation, high implementation costs, algorithmic opacity, cybersecurity risks, and unresolved governance and liability issues. Based on these findings, this study proposes a data-centric and phased research agenda that emphasizes benchmark datasets, human–AI collaboration, lifecycle economic evaluation, explainable AI, and multi-stakeholder governance. The study contributes to the literature by integrating fragmented AI-related research into a structured knowledge map and by clarifying future pathways for developing intelligent, transparent, and resilient construction supply chains. Full article

Deep neural networks (DNNs) have significantly advanced machine perception and reasoning; however, their lack of transparency in decision-making continues to pose a major challenge, particularly in high-stakes domains such as healthcare, finance, and law. This is especially concerning with the black-box nature of convolutional neural networks (CNNs), where the rationale for making a decision can be as important as the decision itself. This paper is driven by a question that is easier to ask than to answer: how can CNNs be made to explain themselves? To answer the question, we wrote a PRISMA-compliant systematic review of 154 studies published between 2017 and 2025. These studies were selected from 4421 studies retrieved through Web of Science, Scopus, IEEE Xplore, and ACM Digital Library. CNN-specific taxonomy was developed. This taxonomy organizes explainable artificial intelligence (XAI) methods on four axes: explanation timing, model dependency, output type, and target component. We found that there is a huge bias in the field regarding post hoc visual methods. Grad-CAM is the most widely cited visual explanation methodology, and within the model-agnostic framework, LIME and SHAP prevail. This research was also the first to analyze standard assessment methods. It was found that out of the 154 studies in the review, 98 used objective methods to evaluate fidelity, stability, or sensitivity. Conversely, fewer than ten of them used human-centered methods to evaluate how tasks were performed, how the users trusted the method, or how the users were prepared to interact with the system. We argue for a dual-reporting convention under which metrics should be reported together at least once, as per the family of metrics. The third contribution is an evidence-based challenge map, where we outline four issues: absence of standardized benchmarks, post hoc mechanism scalability limitations, vulnerability to adversarial perturbations, and the persistent gap between the technical descriptions and human understanding. For each challenge, we propose concrete directions: integrating causal reasoning, adopting participatory evaluation design, and building hybrid transparent architectures. We offer this review as a practical roadmap for researchers and practitioners working toward more explainable deep neural networks. Full article

This study presents a clear and reliable framework for classifying the intent behind scientific citations. It combines multi-model reasoning with concepts from social choice theory. Instead of using a single model, this framework employs three open Large Language Models Gemma, LLaMA, and Mistral. Additionally, we combine their ranked outputs using an exponentially weighted Borda method. By doing so, this approach increases agreement among high-confidence predictions, maintains ranking information, and produces stable, high-quality supervision signals. Consequently, it boosts reliability while remaining transparent. To create a strong experimental basis, we built a large, balanced dataset from the UnarXive corpus, which contains structured full-text scientific articles and citation networks. First, we automatically pulled citation contexts and organized them within a DuckDB-based analytical setup. Then, we rebalanced the dataset across rhetorical categories to enhance representativeness and minimize bias. Finally, we categorized each citation context into one of five roles: background, methodology, comparison, extension, or critique. As a result, the resulting dataset provides a robust foundation for training and evaluation. We trained a SciBERT classifier using these ensemble-generated annotations and tested it on a five-category citation intent classification task. The model achieved a macro F1-score of 0.83, an outstanding result for this type of classification. Indeed, this level of performance shows strong reliability given how challenging it is to differentiate closely related citation functions. Moreover, it demonstrates that combining multiple models produces valuable and distinct supervision signals, capturing subtle rhetorical and semantic patterns that single models often overlook. Furthermore, the framework enhances interpretability. Specifically, the explicit weighting system clarifies how each model contributes to the final outcome. In addition, the deterministic tie-breaking method ensures the outputs are consistent and reproducible. Taken together, these design choices maintain explainability without sacrificing effectiveness. Full article

Nd:S-FAP (Nd³⁺-doped Sr₅(PO₄)₃F) is a high-performance laser material recognized for its large stimulated emission cross-section and broad absorption bands, which are highly desirable for achieving efficient optical gain at the nanoscale. In this work, 5% Nd-doped Nd:S-FAP nanocrystals with an average grain size of approximately 9.2 nm were successfully synthesized via an improved hot-injection method. The results show that the fluorescence lifetime of the nanocrystals is 66 μs (166.56 μs for the ceramic), the quantum yield is 19.7% (69.6% for the ceramic), and the stimulated emission cross-section is 2.29 × 10⁻²⁰ cm² (5.74 × 10⁻²⁰ cm² for the ceramic). These discrepancies are primarily governed by surface-to-volume ratio variation-related surface effects, lattice distortions, and the weakening of non-radiative f–f transition processes due to quantum confinement at the nanoscale. This study reports for the first time the laser optical parameters of Nd:S-FAP nanocrystals, providing an experimental basis for the optimization of precursors for transparent laser ceramics and holding significant importance for the design of laser materials. Full article

Sapphire crystal, owing to its high hardness, chemical inertness, thermal stability, optical transparency, and superior dielectric strength, as well as resistance to scratching, abrasion, friction, and wear, is widely utilized in a broad range of engineering applications. Slicing is the most critical step in sapphire industry processing, as it largely dictates the final surface quality and morphology. Conventional wire sawing methods often lead to undesirable surface defects, while ultrasonic-assisted diamond wire sawing (UADWS) offers potential advantages through enhanced abrasive self-sharpening, micro-hammering, and improved lubricant penetration. However, its influence on sapphire slicing remains insufficiently studied. This study investigates the effects of UADWS parameters—ultrasonic amplitude, horn application position, feed speed, and wire speed—on the surface quality of sapphire crystals. Both single-factor and orthogonal five-level experiments were designed, taking wire and feed speed within industrial parameter ranges. Surface roughness (Ra) and waviness peak–valley (PV) difference were used as evaluation indices, and range and variance analyses were performed. In addition, power regression models were developed to predict Ra and PV under varying conditions. The surface morphology results from single-factor experiments reveal that increasing feed speed and wire speed reduces the effectiveness of ultrasonic assistance, while application horn position exerts only a minor influence. Overall, orthogonal analysis confirmed that the relative influence of process parameters on surface quality follows the order: feed speed > wire speed > amplitude > application horn position. These findings establish a foundation for optimizing the sawing and ultrasonic parameters of UADWS to enhance sapphire surface quality, reduce downstream processing requirements, and clarify the importance of controlling feed speed and wire speed. Full article

Climate change compels the built environment to adopt adaptive Building-Integrated PhotoVoltaic (BIPV) designs balancing energy yields with indoor environmental quality. However, current dynamic facades suffer from algorithmic opacity and an operational bias prioritising generation over human comfort. This study systematically reviews the energy–comfort nexus within intelligent building skins. Following PRISMA 2020 guidelines, we searched nine databases (including Scopus and Web of Science) for empirical studies (2016–early 2026) evaluating dynamic BIPV control. Methodological quality and bias risk were assessed via a rigorous framework evaluating control transparency and simulation validity. From 2423 initial records, 92 high-quality studies were selected and structured into six adaptive BIPV design clusters (e.g., Double-Skin Facades, kinetic shading, semi-transparent glazing). The synthesis reveals a polarisation between thermo-mechanical integration and computational–geometric optimisation. An energy-centric bias persists. Power generation dominates over 92% of research, whereas glare evaluation remains below 20%. Additionally, nearly 60% of studies fail to explicitly define their control parameters. Current evidence limitations include a heavy reliance on idealised 1D simulations, a strong Northern Hemisphere geographical concentration, and a scarcity of robust data on dynamic bifacial systems. Bridging the gap between PV hardware and intelligent control, this study delivers a comprehensive theoretical framework and an actionable roadmap. Full article