Search Results (369)

Accurate staging of bladder cancer is critical for guiding clinical management, particularly the distinction between non–muscle-invasive (T1) and muscle-invasive (T2–T4) disease. Although MRI offers superior soft-tissue contrast, image interpretation remains operator-dependent and subject to inter-observer variability. This study proposes an automated deep learning framework for MRI-based bladder cancer staging to support standardized radiological interpretation. A sequential AI-based pipeline was developed, integrating hybrid tumor segmentation using YOLOv11 for lesion detection and DeepLabV3 for boundary refinement, followed by three deep learning classifiers (VGG19, ResNet50, and Vision Transformer) for MRI-based stage prediction. A total of 416 T2-weighted MRI images with radiology-derived stage labels (T1–T4) were included, with data augmentation applied during training. Model performance was evaluated using accuracy, precision, recall, F1-score, and multi-class AUC. Performance uncertainty was characterized using patient-level bootstrap confidence intervals under a fixed training and evaluation pipeline. All evaluated models demonstrated high and broadly comparable discriminative performance for MRI-based bladder cancer staging within the present dataset, with high point estimates of accuracy and AUC, particularly for differentiating non–muscle-invasive from muscle-invasive disease. Calibration analysis characterized the probabilistic behavior of predicted stage probabilities under the current experimental setting. The proposed framework demonstrates the feasibility of automated MRI-based bladder cancer staging derived from radiological reference labels and supports the potential of deep learning for standardizing and reproducing MRI-based staging procedures. Rather than serving as an independent clinical decision-support system, the framework is intended as a methodological and workflow-oriented tool for automated staging consistency. Further validation using multi-center datasets, patient-level data splitting prior to augmentation, pathology-confirmed reference standards, and explainable AI techniques is required to establish generalizability and clinical relevance. Full article

Water quality monitoring faces increasing challenges due to accelerating industrialization and urbanization, demanding accurate, real-time, and reliable prediction technologies. This study presents a novel ensemble learning framework integrating Gaussian Process Regression, Support Vector Regression, and Random Forest algorithms for high-precision water quality pH prediction. The research utilized a comprehensive spatiotemporal dataset, comprising 11 water quality parameters from 37 monitoring stations across Georgia, USA, spanning 705 days from January 2016 to January 2018. The ensemble model employed a dynamic weight allocation strategy based on cross-validation error performance, assigning optimal weights of 34.27% to Random Forest, 33.26% to Support Vector Regression, and 32.47% to Gaussian Process Regression. The integrated approach achieved superior predictive performance, with a mean absolute error of 0.0062 and coefficient of determination of 0.8533, outperforming individual base learners across multiple evaluation metrics. Statistical significance testing using Wilcoxon signed-rank tests with a Bonferroni correction confirmed that the ensemble significantly outperforms all individual models (p < 0.001). Comparison with state-of-the-art models (LightGBM, XGBoost, TabNet) demonstrated competitive or superior ensemble performance. Comprehensive ablation experiments revealed that Random Forest removal causes the largest performance degradation (+4.43% MAE increase). Feature importance analysis revealed the dissolved oxygen maximum and conductance mean as the most influential predictors, contributing 22.1% and 17.5%, respectively. Cross-validation results demonstrated robust model stability with a mean absolute error of 0.0053 ± 0.0002, while bootstrap confidence intervals confirmed narrow uncertainty bounds of 0.0060 to 0.0066. Spatiotemporal analysis identified station-specific performance variations ranging from 0.0036 to 0.0150 MAE. High-error stations (12, 29, 33) were analyzed to distinguish characteristics, including higher pH variability and potential upstream pollution influences. An integrated software platform was developed featuring intuitive interface, real-time prediction, and comprehensive visualization tools for environmental monitoring applications. Full article

Evaluating artificial intelligence (AI) models in clinical medicine requires more than conventional metrics such as accuracy, Area Under the Receiver Operating Characteristic (AUROC), or F1-score, which often overlook key considerations such as fairness, reliability, and real-world utility. We introduce CUES as a multiplicative composite score for clinical prediction models; it is defined as $\mathrm{CUES}=(C\cdot U\cdot E\cdot S{)}^{1/4}$, where C represents calibration, U integrated clinical utility, E equity across patient subpopulations, and S sampling stability. We formally establish boundedness, monotonicity, and differentiability on the domain ${\left(\mathrm{0,1}\right]}^4$, derive first-order sensitivity relations, and provide asymptotic approximations for its sampling distribution via the delta method. To facilitate inference, we propose bootstrap procedures for constructing confidence intervals and for comparative model evaluation. Analytic examples illustrate how CUES can diverge from traditional metrics, capturing dimensions of predictive performance that are essential for clinical reliability but often missed by AUROC or F1-score alone. By integrating multiple facets of clinical utility and robustness, CUES provides a comprehensive tool for model evaluation, comparison, and selection in real-world medical applications. Full article

Background/Objectives: Peripheral arterial disease (PAD) is a common yet often unrecognized complication in patients receiving peritoneal dialysis (PD). Considering that ankle–brachial index (ABI) can be difficult to interpret in this population, additional vascular biomarkers are needed. Selenoprotein P (SePP) is a major selenium transport protein with antioxidant and metabolic regulatory functions and may reflect vascular stress relevant to PAD. We investigated the association of circulating SePP levels with ABI-defined PAD in patients on PD. Methods: In this cross-sectional analysis of 98 patients on PD, ABI was assessed using an automated oscillometric device, and ABI < 0.9 was defined as ABI-defined PAD. Serum SePP levels were measured using enzyme-linked immunosorbent assay. Results: ABI-defined PAD was identified in 20 patients (20.4%). Compared with patients with normal ABI, those with ABI-defined PAD were older (p = 0.014) and had significantly higher prevalence of diabetes mellitus (p = 0.033), longer PD vintage (p = 0.036), higher fasting glucose (p = 0.005) and C-reactive protein (p = 0.003) levels, and lower SePP concentrations (p < 0.001). Low SePP level remained independently associated with ABI-defined PAD after multivariate adjustment (odds ratio 0.930, 95% confidence interval 0.771–0.997; p = 0.032) and consistently across reinforced bootstrap resampling. SePP correlated positively with ABI on the left (p = 0.001) and right (p = 0.002) sides. Conclusions: Among patients undergoing PD, a low serum SePP level was independently associated with ABI-defined PAD and positively associated with ABI, suggesting that SePP may serve as an associative biomarker reflecting vascular vulnerability rather than a diagnostic indicator in this population. Full article

Background: Irritable Bowel Syndrome (IBS) is a heterogeneous gastrointestinal disorder characterized by complex brain–gut interactions. Patient Similarity Networks (PSNs) offer a novel approach for exploring this heterogeneity and identifying clinically relevant patient subgroups. Methods: We analyzed data from 78 participants (49 IBS patients and 29 healthy controls) with 36 brain morphometric measures (FreeSurfer v7.4.1) and 6 measures of cognitive functions (5 RBANS domain indices plus a Total Scale score). PSNs were constructed using multiple similarity measures (Euclidean, cosine, correlation-based) with Gaussian kernel transformation. We performed community detection (Louvain algorithm), centrality analyses, feature importance analysis, and correlations with symptom severity. Statistical validation included bootstrap confidence intervals and permutation testing. Results: The PSN comprised 78 nodes connected by 469 edges, with four communities detected. These communities did not significantly correspond to diagnostic groups (Adjusted Rand Index = 0.011, permutation $p=0.212$), indicating IBS patients and healthy controls were intermixed. However, each community exhibited distinct neurobiological profiles: Community 1 (oldest, preserved cognition) showed elevated intracranial volume but reduced subcortical gray matter; Community 2 (youngest, most severe IBS symptoms) had elevated cortical volumes but reduced white matter; Community 3 (most balanced IBS/HC ratio, mildest IBS symptoms) showed the largest subcortical volumes; Community 4 (lowest cognitive performance across multiple domains) displayed the lowest RBANS scores alongside high IBS prevalence. Top network features included subcortical structures, corpus callosum, and cognitive indices (Language, Attention). Conclusions: PSN identifies brain–cognition communities that cut across diagnostic categories, with distinct feature profiles suggesting different hypothesis-generating neurobiological patterns within IBS that may inform personalized treatment strategies. Full article

We present an adaptive governance engine for blockchain-enabled Vehicular Ad Hoc Networks (VANETs) that regulates the latency–energy–coherence trade-off under rapid topology changes. The core contribution is an Ideal Information Cycle (an operational abstraction of information injection/validation) and a modular VANET Engine implemented as a real-time control loop in NS-3.35. At runtime, the Engine monitors normalized Shannon entropies—informational entropy S over active transactions and spatial entropy $H_{\mathrm{spatial}}$ over occupancy bins (both on $[0,1]$)—and adapts the consensus mode (latency-feasible PoW versus signature/quorum-based modes such as PoS/FBA) together with rigor parameters via calibrated policy maps. Governance is formulated as a constrained operational objective that trades per-block resource expenditure (radio + cryptography) against a Quality-of-Information (QoI) proxy derived from delay/error tiers, while maintaining timeliness and ledger-coherence pressure. Cryptographic cost is traced through counted operations, $E_{\mathrm{crypto}}=e_h{n}_{\mathrm{hash}}+e_{\mathrm{sig}}{n}_{\mathrm{sig}}$, and coherence is tracked using the LCP-normalized definition $D_{\mathrm{ledger}}\left(t\right)$ computed from the longest common prefix (LCP) length across nodes. We evaluate the framework under urban/highway mobility, scheduled partitions, and bounded adversarial stressors (Sybil identities and Byzantine proposers), using 600 s runs with 30 matched random seeds per configuration and 95% bias-corrected and accelerated (BCa) bootstrap confidence intervals. In high-disorder regimes ($S\gtrsim 0.8$), the Engine reduces total per-block energy (radio + cryptography) by more than 90% relative to a fixed-parameter PoW baseline tuned to the same agreement latency target. A consensus-first triggering policy further lowers agreement latency and improves throughput compared with broadcast-first baselines. In the emphasized urban setting under high mobility ($v=30$ m/s), the Engine keeps agreement/commit latency in the sub-100 ms range while maintaining finality typically within sub-150 ms ranges, bounds orphaning (≤10%), and reduces average ledger divergence below $0.07$ at high spatial disorder. The main evaluation is limited to $N\le 100$ vehicles under full PHY/MAC fidelity. PoW targets are intentionally latency-feasible and are not intended to provide cryptocurrency-grade majority-hash security; operational security assumptions and mode transition safeguards are discussed in the manuscript. Full article

Accurate volatility forecasting is essential for risk management in increasingly interconnected financial markets. Traditional econometric models capture volatility clustering but struggle to model nonlinear cross-market spillovers. This study proposes a Temporal Graph Attention Network (TemporalGAT) for multi-horizon volatility forecasting, integrating LSTM-based temporal encoding with graph convolutional and attention layers to jointly model volatility persistence and inter-market dependencies. Market linkages are constructed using the Diebold–Yilmaz volatility spillover index, providing an economically interpretable representation of directional shock transmission. Using daily data from major global equity indices, the model is evaluated against econometric, machine learning, and graph-based benchmarks across multiple forecast horizons. Performance is assessed using MSE, $R^2$, MAFE, and MAPE, with statistical significance validated via Diebold–Mariano tests and bootstrap confidence intervals. The study further conducts a strict expanding-window robustness test, comparing fixed and dynamically re-estimated spillover graphs in a fully out-of-sample setting. Sensitivity and scenario analyses confirm robustness across hyperparameter configurations and market regimes, while results show no systematic gains from dynamic graph updating over a fixed spillover network. Full article

This study re-evaluates ROC-AUC for binary classification under severe class imbalance (<3% positives). Despite its widespread use, ROC-AUC can mask operationally salient differences among classifiers when the costs of false positives and false negatives are asymmetric. Using three benchmarks, credit-card fraud detection (0.17%), yeast protein localization (1.35%), and ozone level detection (2.9%), we compare ROC-AUC with Matthews Correlation Coefficient, F2-score, H-measure, and PR-AUC. Our empirical analyses span 20 classifier–sampler configurations per dataset, combined with four classifiers (Logistic Regression, Random Forest, XGBoost, and CatBoost) and four oversampling methods plus a no-resampling baseline (no resampling, SMOTE, Borderline-SMOTE, SVM-SMOTE, ADASYN). ROC-AUC exhibits pronounced ceiling effects, yielding high scores even for underperforming models. In contrast, MCC and F2 align more closely with deployment-relevant costs and achieve the highest Kendall’s τ rank concordance across datasets; PR-AUC provides threshold-independent ranking, and H-measure integrates cost sensitivity. We quantify uncertainty and differences using stratified bootstrap confidence intervals, DeLong’s test for ROC-AUC, and Friedman–Nemenyi critical-difference diagrams, which collectively underscore the limited discriminative value of ROC-AUC in rare-event settings. The findings recommend a shift to a multi-metric evaluation framework: ROC-AUC should not be used as the primary metric in ultra-imbalanced settings; instead, MCC and F₂ are recommended as primary indicators, supplemented by PR-AUC and H-measure where ranking granularity and principled cost integration are required. This evidence encourages researchers and practitioners to move beyond sole reliance on ROC-AUC when evaluating classifiers in highly imbalanced data. Full article

The analysis of road traffic accidents often reveals asymmetric patterns, providing insights that support the development of preventive measures, reduce fatalities, and improve road safety interventions. The Rayleigh distribution, a continuous distribution with inherent asymmetry, is well suited for modeling right-skewed data and is widely used in scientific and engineering fields. It also shares structural characteristics with other skewed distributions, such as the Weibull and exponential distributions, and is particularly effective for analyzing right-skewed accident data. This study considers several approaches for constructing confidence intervals, including the percentile bootstrap, bootstrap with standard error, generalized confidence interval, method of variance estimates recovery, normal approximation, Bayesian Markov Chain Monte Carlo, and Bayesian highest posterior density methods. Their performance was evaluated through Monte Carlo simulation based on coverage probabilities and expected lengths. The results show that the HPD method achieved coverage probabilities at or above the nominal confidence level while providing the shortest expected lengths. Finally, all proposed confidence intervals were applied to fatalities recorded during the seven hazardous days of Thailand’s Songkran festival in 2024 and 2025. Full article

This paper develops a quantitative framework for climate–financial risk measurement that combines a spatially explicit jump–diffusion asset–loss model with prudentially aligned risk metrics. The approach connects regional physical hazards and transition variables derived from climate-consistent pathways to asset returns and credit parameters through the use of climate-adjusted volatilities and jump intensities. Fat tails and geographic heterogeneity are captured by it, which conventional diffusion-based or purely narrative stress tests fail to reflect. The framework delivers portfolio-level Spatial Climate Value-at-Risk (SCVaR) and Expected Shortfall (ES) across scenario–horizon matrices and incorporates an explicit robustness layer (block bootstrap confidence intervals, unconditional/conditional coverage backtests, and structural-stability tests). All ES measures are understood as Conditional Expected Shortfall (CES), i.e., tail expectations evaluated conditional on climate stress scenarios. Applications to bank loan books, pension portfolios, and sovereign exposures show how climate shocks reprice assets, alter default and recovery dynamics, and amplify tail losses in a region- and sector-dependent manner. The resulting, statistically validated outputs are designed to be decision-useful for Internal Capital Adequacy Assessment Process (ICAAP) and Pillar 2: climate-adjusted capital buffers, scenario-based stress calibration, and disclosure bridges that complement alignment metrics such as the Green Asset Ratio (GAR). Overall, the framework operationalises a move from exposure tallies to forward-looking, risk-sensitive, and auditable measures suitable for supervisory dialogue and internal risk appetite. Full article

Background/Objectives: To construct and compare multivariable prediction models for the early prediction of large-for-gestational-age (LGA) neonates, using ultrasound biometry and maternal characteristics. Methods: This retrospective cohort study analyzed data from singleton pregnancies that underwent routine ultrasound examinations at 30⁺⁰–34⁺⁰ weeks of gestation. Ultrasound parameters included fetal abdominal circumference (AC), head circumference (HC), femur length (FL), HC-to-AC ratio, mean uterine artery pulsatility index (mUtA-PI), and presence of polyhydramnios. LGA neonates were defined as those having a birthweight > 90th percentile. Logistic regression was used to evaluate associations between ultrasound markers and LGA after adjusting for the following maternal and pregnancy-related covariates: maternal age, body mass index, parity, gestational diabetes mellitus (GDM), pre-existing diabetes, previous cesarean section (PCS), assisted reproductive technology (ART) use, smoking, hypothyroidism, and chronic hypertension. Associations were expressed as adjusted odds ratios (aORs) with 95% confidence intervals (CIs). Three prognostic models were developed utilizing the following predictors: (i) biometric ultrasound measurements including AC, HC-to-AC ratio, FL, UtA-PI, and polyhydramnios (Model 1), (ii) a combination of biometric ultrasound measurements and clinical–maternal data (Model 2), and (iii) only the estimated fetal weight (EFW) (Model 3). Results: In total, 3808 singleton pregnancies were included in the analyses. The multivariable analysis revealed that AC (aOR 1.07, 95% CI [1.06, 1.08]), HC to AC (aOR 1.01, 95% CI [1.006, 1.01]), FL (aOR 1.01, 95% CI [1.009, 1.01]), and the presence of polyhydramnios (aOR 4.97, 95% CI [0.7, 58.8]) were associated with an increased risk of LGA, while a higher mUtA-PI was associated with a reduced risk (aOR 0.98, 95% CI [0.98, 0.99]). Maternal parameters, such as GDM, pre-existing diabetes, elevated pre-pregnancy BMI, absence of uterine artery notching, mUtA-PI, and multiparity, were significantly higher in the LGA group. Both models 1 and 2 showed similar performance (AUCs: 84.7% and 85.3%, respectively) and outperformed model 3 (AUC: 77.5%). Bootstrap and temporal validation indicated minimal overfitting and stable model performance, while decision curve analysis supported potential clinical utility. Conclusions: Models using biometric and Doppler ultrasound at 30–34 weeks demonstrated good discriminative ability for predicting LGA neonates, with an AUC up to 84.7%. Adding maternal characteristics did not significantly improve performance, while the biometric model performed better than EFW alone. Sensitivity at conventional thresholds was low but increased substantially when lower probability cut-offs were applied, illustrating the model’s threshold-dependent flexibility for early risk stratification in different clinical screening needs. Although decision curve analysis was performed to explore potential clinical utility, external validation and prospective assessment in clinical settings are still needed to confirm generalizability and to determine optimal decision thresholds for clinical application. Full article

This study develops approaches for estimating the common percentile of Birnbaum–Saunders (BS) distributions and applies them to daily PM_2.5 concentration data from six monitoring stations in Chiang Mai Province, Thailand. Percentiles provide a robust representation of typical pollutant exposure, being less sensitive to outliers and suitable for skewed environmental data. Estimating the same percentile across multiple monitoring sites offers a standardized metric for regional air quality assessment, enabling meaningful comparisons and informing evidence-based environmental policy. Four statistical approaches—Generalized Confidence Interval (GCI), bootstrap, Bayesian, and Highest Posterior Density (HPD)—were employed to construct confidence intervals (CIs) for the common percentile. Simulation studies evaluated the methods in terms of average length (AL) and coverage probability (CP), showing that the GCI approach offers the best balance between precision and reliability. Application to real PM_2.5 data confirmed that the BS distribution appropriately models pollutant concentrations and that the common percentile provides a meaningful measure for environmental assessment. These findings highlight the GCI method as a robust tool for constructing CIs in environmental data analysis. Full article

While face recognition systems are increasingly deployed in critical domains, they remain vulnerable to presentation attacks and exhibit significant demographic bias, particularly affecting African populations. This paper presents a fairness-aware Presentation Attack Detection (PAD) system using Local Binary Patterns (LBPs) with novel ethnicity-aware processing techniques specifically designed for African contexts. Our approach introduces three key technical innovations: (1) adaptive preprocessing with differentiated Contrast-Limited Adaptive Histogram Equalization (CLAHE) parameters and gamma correction optimized for different skin tones, (2) group-specific decision threshold optimization using Equal Error Rate (EER) minimization for each ethnic group, and (3) three novel statistical methods for PAD fairness evaluation such as Coefficient of Variation analysis, McNemar’s significance testing, and bootstrap confidence intervals representing the first application of these techniques in Presentation Attack Detection. Comprehensive evaluation on the Chinese Academy of Sciences Institute of Automation-SURF Cross-ethnicity Face Anti-spoofing dataset (CASIA-SURF CeFA) dataset demonstrates significant bias reduction achievements: a 75.6% reduction in the accuracy gap between African and East Asian subjects (from 3.07% to 0.75%), elimination of statistically significant bias across all ethnic group comparisons, and strong overall performance, with 95.12% accuracy and 98.55% AUC. Our work establishes a comprehensive methodology for measuring and mitigating demographic bias in PAD systems while maintaining security effectiveness, contributing both technical innovations and statistical frameworks for inclusive biometric security research. Full article

Objective: To investigate long-term prognosis and impact factors in children with vasovagal syncope (VVS) receiving metoprolol therapy. Methods: This retrospective study included children with VVS who underwent metoprolol therapy at the Pediatric Syncope Unit of Peking University First Hospital between January 2012 and November 2023. Baseline demographic data, pre-treatment indices, including head-up tilt test (HUTT) and 24 h Holter monitoring, were collected. All participants received standardized metoprolol therapy for a minimum duration of one month. Follow-up was conducted between June and July 2025, with syncope recurrence as the primary endpoint. Multivariable Cox proportional hazards regression analysis was performed to identify independent impact factors of prognosis and to construct a Prognostic Risk Score (PRS) model. The model’s performance was rigorously validated through receiver operating characteristic (ROC) curve analysis, decision curve analysis (DCA), and Bootstrap resampling (1000 iterations). Furthermore, children were stratified into high- and low-risk groups based on median PRS values. Kaplan–Meier survival analysis was then performed to assess the model’s discriminative efficacy. Results: This study included 97 children diagnosed with VVS. The median duration of metoprolol therapy was 2.5 months (interquartile range [IQR]: 2.0–3.0 months), with a median follow-up period of 59 months (IQR: 25.5–72 months). During follow-up, syncope recurrence was observed in 37 patients, while 60 patients remained symptom-free. COX regression analysis showed that time-domain indices of heart rate variability (HRV), including the standard deviation of all NN intervals (SDNN) and the triangular index (TR), as well as the frequency-domain index of HRV very low frequency (VLF), were relative factors of the long-term prognosis in children with VVS treated with metoprolol. Based on the above three identified factors, the PRS model was calculated as: PRS = 0.03 × SDNN − 0.02 × VLF − 0.1 × TR. ROC showed that the area under the curve (AUC) for discriminative power related to long-term prognosis was 0.808 (p < 0.01). The cumulative recurrence rate of symptoms in the high-risk score group was significantly higher than that in the low-risk score group (p < 0.01). The DCA curve demonstrated the clinical applicability of the model. Bootstrap internal verification indicated high stability, with the bias-corrected and accelerated (Bca) confidence interval (CI) of the C index ranging from 0.71 to 0.89. Conclusions: After metoprolol treatment, 38.1% of children with VVS experienced syncope recurrence during a median follow-up period of 59 months. Baseline HRV index, SDNN, TR, and VLF were identified as factors associated with the long-term prognosis of children with VVS treated with metoprolol. The PRS model based on the above indices demonstrated good value in linking to the individual long-term prognosis. Full article

To address the challenge of simultaneously modelling temporal evolution and static properties in fatigue life prediction, this paper proposes a Hybrid GRU–Attention–DNN model: The Gated Recurrent Unit (GRU) captures time-evolution features, while the attention mechanism adaptively focuses on critical stages. These are then fused with static properties via a fully connected network to generate life estimates. Training and validation were conducted using an 8:2 split, with baselines including Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and GRU. Performance was evaluated using the coefficient of determination (R²), root mean squared error (RMSE), mean absolute error (MAE), and root mean squared logarithmic error (RMSLE), together with error band plots. Results demonstrate that the proposed model outperforms baseline CNN/GRU/LSTM models in overall accuracy and robustness, and that these improvements remain statistically significant according to bootstrap confidence intervals (CI) of R², RMSE, MAE and RMSLE on the test set. Additionally, this paper conducts an interpretability analysis: attention visualisations reveal the model’s significant emphasis on the early stages of the lifespan. Time window masking experiments further indicate that removing early information causes the most significant performance degradation. Both lines of evidence show high consistency in qualitative and quantitative trends, providing a basis for engineering sampling window design and trade-offs in test duration. Full article