Search Results (5,301)

Sustainable groundwater management is essential for water security and human health protection. Fluoride contamination is a serious concern for the sustainable drinking water supply in many parts of Pakistan, including Balochistan, where arid climate conditions and geological formations support the enrichment of fluoride. The toxic nature of fluoride contamination has resulted in negative health impacts on the local population. Conventional geostatistical techniques are usually ineffective to delineate the nonlinear relationships that affect the distribution of fluoride. This study aims to develop a machine learning-driven spatial modelling framework for classifying the spatial distribution of fluoride contamination in groundwater across the study area. The model will help to understand the spatial variability of fluoride contamination and its controlling factors, essential for effective mitigation and early warning systems. Physiochemical elements were used as predictive features in this study, utilizing a unified feature importance framework combining hydrogeochemical analysis, spatial distribution assessment, and ensemble SHAP-based interpretation to identify consistent predictors. Model performance was evaluated using a nested cross-validation framework, followed by validation on an independent geology-informed spatial holdout test set to ensure realistic generalization. Among machine learning models, the Logistic Regression (LR), Support Vector Classifier (SVC), XGBoost (XGB), Decision Tree (DT), Gaussian Naïve Bayes (GNB), and K-Nearest Neighbours (KNN) were evaluated. Support Vector Classifier (SVC) demonstrated a high predictive performance. On the independent spatial holdout dataset, SVC achieved an overall accuracy of 0.75 and an area under the receiver operating characteristic curve (AUC) of 0.821. In addition to classification, a human health risk assessment was conducted using chronic daily intake (CDI) and hazard quotient (HQ) calculations for children and adults, identifying several high-risk water supply schemes. The prediction maps successfully delineated high-risk fluoride points across specific areas, offering a tool for sustainable groundwater management. This study helps to achieve a Sustainable Development Goal (Clean Water and Sanitation, SDG#6) and promotes long-term sustainable planning in water-stressed areas by integrating spatial machine learning mapping and health risk assessment. Full article

Background/Objectives: This study aims to develop a comprehensive Clinical Decision Support System (CDSS) that integrates multi-sequence sacroiliac joint (SIJ) MRIs with rheumatological, clinical, and laboratory findings into the decision-making process for the early diagnosis of axial spondyloarthritis (axSpA), incorporating segmentation-supported explainability. Methods: Multi-sequence SIJ MRI data (T1-WI, T2-WI, STIR, and PD-WI) were analysed from 367 participants (n = 193 axSpA; n = 174 non-axSpA controls). Sequence-based classification was performed using VGG16, ResNet50, DenseNet121, and InceptionV3 models; additionally, a lightweight and parameter-efficient SacroNet architecture was developed. Slice-level probability scores were converted to patient-level scores using the Dynamic Top-K Averaging method. Image-based scores were combined with a logistic regression-based clinical risk score using weighted linear integration (0.60 image/0.40 clinical) and a conservative threshold (τ = 0.70). Grad-CAM was applied for visual interpretability. Furthermore, to support the diagnostic outcomes with precise spatial data, active inflammation in STIR and T2-WI sequences was segmented. For this purpose, the MDC-UNet model was employed and compared with baseline U-Net derivatives. Results: Sequence-specific analysis showed VGG16 performing best on T1-WI (AUC = 0.920; Accuracy = 0.878) and DenseNet121 on STIR (AUC = 0.793; Accuracy = 0.771). The SacroNet architecture provided competitive classification performance at the patient level despite its low number of parameters (~110 K). Furthermore, MDC-UNet successfully segmented active inflammation, yielding Dice scores of 0.752 (HD95: 19.25) for STIR and 0.682 (HD95: 26.21) for T2-WI. Conclusions: The findings demonstrate that patient-level decision integration based on multi-sequence MRI, when used in conjunction with clinical risk scoring and segmentation-assisted interpretability, can provide a feasible and interpretable DSS framework for the early diagnosis of axSpA. Full article

This paper examines the relationship between artificial intelligence (AI) innovation and bank performance, the organizational channels through which these relationships operate, and the role of firm-wide adoption in shaping outcomes. Using patent-based measures of AI innovation for 31 large U.S. commercial banks from 2015 to 2024 based on the Federal Reserve’s Large Bank classification and employing panel regressions with bank and year fixed effects, we find that AI innovation is associated with improved asset quality but higher operating costs and lower profitability in the short run. Our two-step mediation analysis implies that AI innovation induces organizational changes through diminishing employee scale and branch networks, which mitigates management efficiency and profitability. Importantly, firm-wide AI adoption mitigates the adverse association between AI innovation and both management and profitability prior to adoption, suggesting that the realization of AI’s benefits requires organizational adaptation and coordinated deployment. Dynamic tests further support the productivity “J-curve” of AI innovation. Our findings suggest that bank managers should align AI investment with organizational restructuring and coordinated deployment, while regulators should account for short-term adjustment costs when evaluating the performance implications of AI adoption. Full article

Background: Patients with estrogen receptor (ER)-positive ductal carcinoma in situ (DCIS) derive a greater benefit from endocrine therapy than patients with ER-negative disease. The relevance of ER status and progesterone receptor (PR) status in DCIS to radiation therapy has not been well explored. Methods: Patients undergoing breast-conserving surgery with or without radiation were grouped by ER and PR status and matched using rank-based Mahalanobis optimal matching with respect to lesion size and grade and patient age and race. Cumulative incidences were estimated and competing risk regressions with subdistribution hazard ratios (sHRs) were calculated. Results: Among patients who underwent breast-conserving surgery only, 369 patients with ER-PR- disease were matched to 738 patients with ER+PR+ disease (1:2 matching). In multivariate models, patients with ER-PR- disease were at increased risk of any invasive events (sHR = 2.47, p = 0.007) and early ipsilateral invasive events (sHR = 2.64, p = 0.02 in the 0-to-4-year period) relative to patients with ER+PR+ disease. Among patients who underwent breast-conserving surgery with adjuvant radiation, 1498 patients with ER+PR+ disease were matched to 1498 patients with ER-PR- disease. No significant differences were noted with respect to cumulative incidence of any invasive event (5.6% vs. 5.6%) or ipsilateral invasive events (1.9% vs. 2.9%). In multivariate models, no significant differences were noted. Patients with ER-PR+ lesions had similar cumulative incidences of ipsilateral invasive events to patients with ER-PR- disease in the absence of radiation (5.9% vs. 5.9%) and similar cumulative incidences of contralateral invasive events to patients with ER+PR+ disease when radiation was administered (3.2% vs. 4.2%). Conclusion: The statuses of ER and PR carry independent prognostic and therapeutic implications beyond those of traditional clinicopathologic risk factors. Given that ER and PR statuses are routinely collected for patients with DCIS, incorporation of these variables into clinicopathologic risk classification systems is warranted. Full article

Satellite remote sensing offers a cost-effective solution for the continuous monitoring of black and odorous water bodies (BOWs). However, limitations in spatial and spectral resolution hinder the quantitative inversion of water quality parameters and the precise assessment of risk levels using satellite data alone. To address this challenge, this study proposes a synergistic approach combining satellite and Unmanned Aerial Vehicle (UAV) remote sensing to rapidly identify potentially polluted water bodies and quantitatively assess their risk levels. First, a Black and Odorous Water Index (MBOWI) was constructed based on reflectance characteristics in the visible to near-infrared bands to screen for potential black and odorous water bodies using satellite imagery. Subsequently, high-resolution multispectral UAV imagery, integrated with in situ sampling data, was employed to develop machine learning models for inverting key water quality parameters, including Chemical Oxygen Demand (COD), Dissolved Oxygen (DO), Total Phosphorus (TP) and Ammonia Nitrogen (NH₃-N). Comparative analysis of Polynomial Regression (PR), Random Forest (RF), and Simulated Annealing-optimized Support Vector Regression (SA-SVR) revealed that RF and SA-SVR exhibited superior performance in inverting four non-optically active water quality parameters due to their robust nonlinear fitting capabilities, with the mean Adjusted Coefficient of Determination ($R_{adj}^2$) ranging from 0.57 to 0.69. Water quality classification based on the single-factor worst-case method achieved an overall accuracy of 0.70 across validation samples. Notably, for Class V (heavily polluted) water bodies, both classification accuracy and recall rate reached 0.89, demonstrating the model’s high precision in identifying high-risk waters. Finally, the proposed framework was applied to northern Zhejiang Province to assess seven potential black and odorous water bodies, successfully identifying four as high-risk and one as low-risk. This study validates satellite and UAV synergistic remote sensing for the hierarchical risk management of black and odorous water bodies. Full article

Lightweight day-ahead photovoltaic (PV) forecasting models encounter a significant technical challenge: under resource-constrained deployment conditions, it is difficult to simultaneously address weather-regime heterogeneity, maintain model interpretability, and preserve adaptability as operating conditions evolve. To address this issue, we propose a multimodal short-term photovoltaic (PV) forecasting method that integrates weather-mode partitioning using the Clustering Using Representatives (CURE) algorithm with a self-updating Batch-Lasso model. First, the meteorological-PV dataset is partitioned along two dimensions by combining seasonal grouping with CURE clustering within each season, producing representative weather modes and enhancing the fidelity of weather pattern classification. Second, to extract informative predictors from high-dimensional meteorological inputs while maintaining interpretability, we formulate per-mode Lasso regression and adopt the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA) to efficiently solve for the sparse regression coefficients. Third, we introduce a batch-based self-update and correction mechanism with rollback verification, enabling the mode-specific models to be refreshed as new historical data become available while preventing performance degradation. Compared with representative machine learning baselines, the proposed method maintains competitive accuracy with substantially lower computational and storage overhead, enabling high-frequency and energy-efficient inference on resource-constrained terminals, thereby reducing operational burdens and computational energy costs and better meeting the deployment needs of sustainable energy systems under heterogeneous weather conditions. Full article

Synthetic Aperture Radar (SAR) enables all-weather maritime surveillance, yet ship-oriented bounding box (OBB) detection remains challenging in complex scenes. Strong sea clutter and dense harbor scatterers often mask the slender characteristics of ships as well as the weak responses of small ships. Meanwhile, the periodicity of angle parameterization introduces regression discontinuities, and near-symmetric, bright-scatterer-dominated signatures further cause heading ambiguity, undermining the stability of orientation prediction. Moreover, in most detectors, multi-scale feature fusion and angle estimation lack explicit coordination, and rotated-box localization performance is often jointly affected by feature degradation and unstable orientation prediction. To this end, we propose a unified framework that simultaneously strengthens multi-scale representations and stabilizes orientation modeling. Specifically, we design a Gaussian-Guided Stage-Aware Deformable Feature Pyramid Network (GSDFPN) and a Coarse-to-Fine Unit-Circle Resolver (CF-UCR). GSDFPN enhances multi-scale fusion with two plug-in components: (i) a Gaussian-guided High-level Semantic Refinement Module (GHSRM) that suppresses clutter-dominated semantics while strengthening ship-responsive cues, and (ii) a Stage-aware Deformable Fusion Module (SDFM) for low-level features, which disentangles channels into a geometry-preserving spatial stream and a clutter-resistant semantic stream, and couples them via deformable interaction with bidirectional cross-stream gating to better capture the inherent slender characteristics of ships and localize small ships. For orientation, CF-UCR decomposes angle prediction into direction-cluster classification and intra-cluster residual regression on the unit circle, effectively mitigating periodicity-induced discontinuities and stabilizing rotated-box estimation. On SSDD+ and RSDD, our method achieves AP/AP${}_{50}$/AP${}_{75}$ of 0.5390/0.9345/0.4529 and 0.4895/0.9210/0.4712, respectively, while reaching AP${}_{s75}$/AP${}_{m75}$/AP${}_{l75}$ of 0.5614/0.8300/0.8392 and 0.4986/0.8163/0.8934, evidencing strong rotated-box localization across target scales in complex maritime scenes. Full article

With the rapid development in data mining and learning analytics, integrating predictive analytics into educational data has become increasingly critical for supporting students’ learning trajectories. In many schooling systems, the academic tracks (such as Liberal Arts or Science) and the performance of junior high school students can substantially shape their subsequent university pathways and career planning. Despite the long-term impact of these decisions, academic track selections and the evaluation of students’ potential are often made without systematic and evidence-based guidance. Predictive computer applications can assist, but the training of accurate models and the selection of adequate features remain key challenges. This paper details our process of engineering such an application comprising two tasks based on 1357 real-world junior high school academic performance records. The first task applies a classification approach to predict students’ academic track orientation, while the second task employs a multi-output regression model to forecast students’ future academic performance in senior high school. Our approach shows that the stacking ensemble model achieved a classification accuracy of 85.76%, whereas the Bi-LSTM model with multi-head attention attained an overall R² exceeding 82% in performance forecasting; both models demonstrated strong and reliable predictive capability. Moreover, the proposed approach provides inherent interpretability by decomposing predictions at the subject level. Feature importance analysis reveals how different academic subjects contribute variably to both academic track decisions and future academic performance, offering actionable insights for academic counselling and future study planning. By bridging predictive modelling with students’ educational and career planning needs, this study advances the practical application of educational data mining and provides support for evidence-based academic guidance and future career choices in real-world contexts. Full article

Operational agricultural monitoring in the Central European lowlands requires timely parcel boundaries; however, unmarked field edges produce minimal spectral contrast in single-date imagery. Previous works demonstrated that harmonic NDVI encoding enables zero-shot field delineation using foundational models, but the influence of the spectral index choice on temporal boundaries remained unquantified. This study systematically evaluates eleven vegetation indices—NDVI, GNDVI, NDRE, EVI, EVI2, SAVI, MSAVI, NDWI, CIg, CIre, and NDYVI—within a fixed harmonic phenology encoding pipeline. A one-year PlanetScope time series (15 × 15 km, Slavonija, Croatia) was decomposed via annual sinusoidal regression to extract per-pixel phase, amplitude, and mean parameters. These harmonic descriptors were mapped to HSV colour channels and segmented using the Segment Anything Model without fine-tuning. Official agricultural parcels (PAAFRD, 2025) provided ground truth for pixel-wise, object-wise, and size-stratified evaluation. Performance stratified into three tiers based on object-wise metrics. Soil-adjusted and enhanced-greenness indices (MSAVI, EVI, EVI2, and SAVI) achieved F1 = 0.51–0.52, and mIoU = 0.70–0.71, statistically outperforming standard ratio formulations (NDVI: F1 = 0.49) and chlorophyll indices (CIg, CIre: F1 = 0.45–0.47). Pixel-wise scores remained compressed (F1 > 0.88 across all indices), indicating consistent interior coverage but index-dependent boundary precision. Error analysis revealed scale-dependent patterns: merging dominated small parcels (<10,000 m²), while fragmentation increased with parcel size. Results demonstrate that spectral formulation is a systematic design factor in phenology-based delineation, with soil background correction and dynamic range compression improving seasonal trajectory separability. The harmonic parameters generated by this framework provide feature-ready input for crop classification, suggesting that integrated boundary extraction and crop mapping workflows merit further investigation. Full article

Background: Reliable identification remains a cornerstone of forensic investigations, particularly when encountering degraded remains or suboptimal biological evidence. This study evaluates the potential of dentine metabolomics, utilizing proton nuclear magnetic resonance (¹H-NMR) spectroscopy, to detect cancer-associated metabolic signatures in dental tissues for forensic applications. Methods: Forty-four non-carious second molars were analyzed, comprising 22 samples from deceased individuals with a documented history of cancer and 22 age- and sex-matched controls. Metabolomic profiling was conducted using ¹H-NMR spectroscopy to identify and quantify dentine metabolites. Statistical evaluation included unsupervised principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), receiver operating characteristic (ROC) curve analysis, and exploratory binary logistic regression. Results: Among the 209 identified metabolites, inosinic acid and 2-ketobutyric acid were identified as the most robust discriminative biomarkers across both multivariate and univariate frameworks. The exploration within-sample predictive model achieved a Nagelkerke R² of 0.822 and an overall classification accuracy of 90.9%, with a specificity of 95.5% and a sensitivity of 86.4%. These key metabolites are fundamentally associated with purine metabolism and oxidative stress pathways frequently dysregulated in oncogenesis. Conclusions: This pilot study suggests that dentine may retain metabolomic information associated with cancer comorbidity under heterogeneous postmortem conditions. However, the findings remain exploratory and require validation in larger cohorts with standardized postmortem variables before practical forensic implementation. Full article

To address traditional beer detection limitations, this study proposes a rapid NIRS-based method for detecting key indicators and verifying authenticity. Designing Single-task (STL) and Multi-task learning (MTL) strategies, it employs Variable Importance in Projection for wavelength selection. Deep spectral features were extracted utilizing a Multi-Head Attention (MHA)-fused Convolutional Neural Network (CNN-MHA), Long Short-Term Memory (LSTM-MHA), and hybrid CNN-LSTM-MHA networks. To further enhance model performance, the Bayesian Optimization Algorithm globally optimized network hyperparameters in STL, alongside hyperparameters and multi-task loss weights in MTL. Partial least squares regression, support vector machine regression, and partial least squares discriminant analysis models were established using these features. Results indicate that the MTL-based CNN-LSTM-MHA network effectively learns shared features across multiple tasks, significantly improving model generalization. Specifically, the coefficients of determination ($R^2$) for alcohol content and original wort concentration in the validation set were 0.996 and 0.997, respectively, with relative root mean square errors (rRMSE) of 2.024% and 2.515%. In the independent test set, the $R^2$ were 0.995 and 0.991, with rRMSE of 2.515% and 2.087%, respectively. Furthermore, 100% classification accuracy was achieved across all datasets. This method provides an efficient technical solution for beer market regulation and real-time detection in production processes. Full article

Background/Objectives: This study evaluated whether linear cephalometric measurements show systematic differences in their central values across birth cohort groups in adults from a clinical population and analyzed the implications of these differences for clinical interpretation when norms and clinical deviations are used as a reference framework. Methods: A cross-sectional observational analytical study was conducted based on 604 lateral cephalometric radiographs of adult patients. Eleven linear cephalometric measurements were obtained and compared across predefined birth cohort groups (<1980, 1980–1989, and 1990–1999) using robust estimators of central tendency through median regression models adjusted for sex, age group, and sagittal skeletal classification. Results: Several linear cephalometric measurements revealed different central values between the birth cohorts, even after adjusting for relevant covariates. Cranial length, anterior cranial base length, posterior facial height, and posterior cranial base length had lower adjusted median values in the 1990–1999 cohort than in the <1980 cohort. The effective maxillary length and maxillary length also differed between cohorts. Mandibular measurements, including mandibular length, corpus length, and ramus height, showed the largest adjusted median contrasts between cohorts. These cohort-associated differences were not uniform across all measurements. Conclusions: Routinely used linear cephalometric measurements present different central values across adult birth cohort groups under comparable clinical conditions. The relative position of a cephalometric value within its reference distribution may vary by birth cohort. This suggests that using fixed reference means and standard deviations could lead to systematic misestimation in adults from various birth cohorts. Cohort-aware interpretation is valuable in routine cephalometric assessments. Full article

Background/Objectives: Risk stratification in asymptomatic carotid stenosis has traditionally relied on the degree of luminal narrowing; however, plaque vulnerability may better predict cerebrovascular events. Ipsilateral silent brain lesions (SBLs) are considered surrogate markers of stroke risk. This study aimed to identify carotid plaque features on duplex ultrasound (DUS) and computed tomography angiography (CTA), as well as circulating biomarkers, associated with ipsilateral SBL in patients with clinically asymptomatic ≥70% internal carotid artery stenosis. Methods: This prospective observational study with cross-sectional imaging analysis included 316 clinically asymptomatic patients with ≥70% carotid stenosis treated between January 2022 and October 2024. All patients underwent cranial non-contrast CT for SBL detection, DUS plaque characterization (according to the Gray–Weale classification and plaque surface morphology), and CTA analysis, including plaque surface, composition, length, and attenuation values categorized according to Schroeder’s criteria (<50 HU lipid-rich; 51–120 HU fibrous; >120 HU calcified). Demographic, clinical, and laboratory parameters, including inflammatory biomarkers, were recorded. Multivariate logistic regression was performed to identify independent predictors of SBL. Results: SBL were detected in 72 patients (22.8%). On DUS, SBL were significantly associated with Gray–Weale class II plaques, heterogeneous composition, and irregular or ulcerated surfaces (all p < 0.001). On CTA, lipid-rich plaques (<50 HU), ulcerated surfaces, heterogeneous morphology, and lower median plaque density were significantly more frequent in the SBL group (all p < 0.001). In multivariate analysis, independent predictors of SBL were male sex (OR 2.2; 95% CI 1.2–5.7; p = 0.029), Gray–Weale class II plaques (p = 0.002), lipid-rich plaque morphology (OR 21.39; 95% CI 6.86–66.76; p < 0.001), and ulcerated plaque surface on CTA (OR 20.62; 95% CI 7.37–57.68; p < 0.001). Conclusions: Specific ultrasound and CT plaque characteristics were associated with ipsilateral silent brain lesions in patients with asymptomatic ≥70% carotid stenosis. A multiparametric imaging approach may improve risk stratification beyond stenosis severity alone. Full article

Introduction: Interpregnancy interval (IPI) plays a critical role in neonatal health, yet optimal spacing remains controversial. This study assessed neonatal outcomes across short and long IPI using three complementary classification approaches to identify consistent patterns of risk. Materials and Methods: In this retrospective cohort study, medical records of 1194 women with a prior live birth who delivered singleton pregnancies in 2024 at a tertiary referral center were analyzed. IPI was calculated as the delivery-to-conception interval (LMP + 14 days). Three IPI classification systems were applied: (1) classical cut-offs (<6, 6–11, 12–23, 24–59, and ≥60 months), (2) quartiles, and (3) tertiles. Primary outcomes included preterm birth, low birth weight (LBW), and NICU admission. Multivariable logistic regression models adjusted for maternal age, gravidity, and previous cesarean delivery. Results: Short IPI (6–11 months) demonstrated the highest NICU admission rates (29.4%). Very long IPI (≥60 months) showed the highest prevalence of LBW (16.6%). Multivariable regression analysis revealed that intervals ≥ 24 months were independently protective against preterm birth (24–59 months: aOR 0.48, p = 0.002; ≥60 months: aOR 0.58, p = 0.042), while maternal age increased preterm birth risk by 7% per year. Short IPI (6–11 months) and very long IPI (≥60 months) independently increased NICU admission risk (aOR 2.29, p = 0.002 and aOR 1.61, p = 0.036, respectively). Previous cesarean delivery was an independent predictor of NICU admission (aOR 1.35; p = 0.048). Conclusions: Short and very long IPIs are associated with increased neonatal morbidity, particularly NICU admission, while the apparent preterm risk in long intervals is largely mediated by maternal age. Once adjusted, IPIs exceeding 24 months demonstrate protective effects against preterm birth. However, the rising trend toward LBW and NICU admission in intervals beyond 5 years suggests that birth-spacing counseling targeting an optimal window of 18–24 months provides the best balance in minimizing competing neonatal risks. Full article

In this paper, we present an extensive evaluation of tabular data distillation methods for downstream classification and regression tasks. Our analysis considers multiple distillation approaches that are problem-type independent (i.e., unsupervised). For downstream learners, we focus on non-neural models such as Random Forest, XGBoost, and Support Vector Machines, as our goal is to evaluate the quality of the distilled data independently of the learner. The evaluation is conducted on 17 classification and nine regression problems. Our findings can be summarized as follows: (1) in all cases, applying a distillation method leads to a decrease in performance compared to the baseline; (2) overall, coreset-based methods are the most effective, with performance losses that are minimal—ranging from around 3% in classification accuracy or regression correlation to, in some cases, being negligible; (3) performance loss is moderately correlated with dataset tailness, measured as the proportion of outliers; (4) all distillation methods alter dataset consistency, narrowing the range of hyperparameter values that yield good performance; and (5) the Coreset Leverage Score remains fast, regardless of the size of the original set and of the distilled set. Full article