Search Results (3,665)

Machine learning (ML), including deep learning, kernel-based classifiers, and ensemble methods, is increasingly used to support clinical diagnosis in medical imaging, biosignal interpretation, and electronic health record (EHR)-based decision support. Despite rapid progress, many diagnostic AI studies still rely on limited retrospective evaluation and single summary measures (e.g., accuracy or AUC), creating a gap between reported model performance and evidence required for safe clinical adoption. This review proposes TRIAGE, a clinically grounded evaluation framework designed to organize diagnostic AI testing as an evidence pipeline aligned with real clinical use cases (screening, triage, second reading, and confirmatory testing). We summarize core discrimination metrics derived from the confusion matrix (sensitivity, specificity, predictive values, likelihood ratios, diagnostic odds ratio, and F-scores) and highlight the importance of prevalence and spectrum effects for interpreting predictive value and clinical workload. We further review evaluation strategies for multi-class and multi-label diagnostic tasks using appropriate aggregation methods (micro, macro, and weighted averaging) and set-based measures such as Hamming loss, exact match ratio, and Jaccard/IoU. Because diagnostic deployment is threshold-dependent, we integrate representation curves (ROC, precision–recall, lift, and cumulative gain) with calibration assessment and clinical utility analysis, including calibration slope, Brier score, and decision-curve analysis. We also address robustness and fairness evaluation, leakage-resistant validation designs (patient-grouped splits, stratified and temporal validation, and external validation), computational constraints relevant to deployment (latency, throughput, and energy use), and statistically sound model comparison with multiplicity control. A structured TRIAGE checklist table summarizing the evaluation parameters described in this review is provided in the main text to support reproducible and clinically interpretable reporting. Full article

Background: Early diagnosis of rheumatoid arthritis (RA) remains challenging due to the limited performance of existing serum biomarkers. This exploratory study aimed to identify novel serum metabolite and lipoprotein biomarkers for RA and to develop interpretable machine learning models for screening. Methods: This study employed ¹H-NMR metabolomics to analyze serum from 77 RA patients and 70 healthy controls, quantifying 38 endogenous metabolites and 112 lipoprotein parameters. Seven key biomarkers were identified using multiple criteria and Least Absolute Shrinkage and Selection Operator (LASSO) regression. The dataset was split into training and testing sets (7:3 ratio), and four machine learning models were constructed. The Random Forest (RF) model was further interpreted using the SHapley Additive exPlanations (SHAP) method. Results: The selected biomarkers, including formic acid and High-density lipoprotein 4 phospholipids (H4PL), showed significant associations with RA. In the internal test set, the RF model demonstrated promising discriminatory ability. Additionally, a proof-of-concept regression model for predicting the Disease Activity Score in 28 joints (DAS-28) score was developed, explaining a portion of its variance (R² = 0.548) in this cohort. Conclusions: This exploratory, single-center study identifies a novel panel of potential biomarkers for RA and provides a preliminary, interpretable predictive tool. The findings, particularly the internally validated high performance of certain markers, are hypothesis-generating and underscore the need for validation in larger, multi-center cohorts. The DAS-28 prediction model also warrants further investigation. Full article

Accurate grading of Oil Palm Fresh Fruit Bunches (FFB) is pivotal for maximizing agricultural yield, yet manual assessment in unstructured environments remains labor-intensive and subjective. While Convolutional Neural Networks (CNNs) offer an automated solution, the conventional strategy of scaling network depth often yields diminishing returns or overfitting on moderately sized datasets. To overcome these limitations, this study proposes the Deep Feature Concatenation (DFC) framework. Rather than deepening a single architecture, this methodology synergizes the spatial hierarchy preservation of ResNet50 with the dense feature-reuse mechanisms of DenseNet121. This fusion creates a composite representation space that captures complementary inductive biases. To ensure computational efficiency, the framework decouples representation learning from inference. Principal Component Analysis (PCA) retains 99% of explained variance while compressing features by 68%. These optimized representations are classified using shallow linear probes. Validated on a single-source dataset expanded to 4000 images (derived from 466 original samples) using a rigorous “Parent–Child” split to prevent data leakage, DFC achieved a peak accuracy of 97.75%. McNemar’s statistical test indicated that this performance outperforms the ResNet50 baseline ($p=0.039$) for SVM classifiers. However, it is critical to note that these results represent a proof of concept based on a limited biological sample size, particularly for rare defect classes. While the model achieved 100% detection accuracy for critical defects within the specific validation set, the high synthetic-to-original ratio necessitates cautious interpretation regarding external validity. This framework provides a practical foundation for future research into high-precision, low-latency grading systems, but multi-center validation on larger, independent datasets is required to confirm broad generalizability across diverse plantation environments. Full article

Photovoltaic (PV) systems nowadays represent an essential component of renewable energy production. However, undetected faults often compromise their reliability, leading to significant energy losses and high maintenance costs. Existing deep learning approaches for PV fault diagnosis have achieved high accuracy, but they require massive, labeled datasets and high computational resources, which make them unsuitable for real-time applications. This paper proposes a lightweight, self-supervised hybrid learning framework for real-time PV fault diagnosis to address these limitations. First, the dataset is split into training, testing, and validation subsets. Thereafter, weighted class calculation steps are performed to overcome the issue of imbalance in the data. Then, a self-supervised pre-training phase is established to enable the encoder to produce effective internal representations prior to the implementation of a supervised fine-tuning classifier, characterized as a lightweight feed-forward network (Dense–Dropout–Dense Softmax), which will be trained using categorical cross-entropy and fault-type labels. Finally, a supervised fine-tuning stage is employed based on the pre-trained hybrid CNN–transformer encoder to perform PV fault classification. The experimental results indicate that the proposed approach outperforms existing models by achieving an overall accuracy of 99.8%, a recall of 99.6%, and an outstanding specificity of 100%. The confusion matrix demonstrates that classification is excellent on all operating types. Runtime analysis indicates that the model processes each sample in 2.78 ms and requires 0.07 MB to store weights of 19,429 parameters, confirming its suitability for real-time deployment. These findings highlight that using a hybrid CNN–Transformer encoder with self-supervised learning can improve fault detection and classification performance while significantly reducing inference time, making it an effective and efficient solution for intelligent PV system monitoring. Full article

Major depressive disorder requires scalable, low-burden screening tools. We examined whether three-channel resting-state EEG can support reliable discrimination between major depressive disorder and healthy controls using a lightweight model compatible with portable implementations. This work makes three main contributions: (i) a compact hybrid fusion model combining raw-window Conv1D embeddings with per-channel spectral–statistical descriptors for three-channel resting-state EEG, (ii) a leakage-resistant subject-independent (cross-subject) evaluation protocol with subject-level inference via majority voting, and (iii) a preliminary external feasibility test on an independent portable three-channel cohort without fine-tuning. The proposed model fuses a Conv1D encoding of raw ≈15 s eyes-closed windows (3840 samples; 15.36 s at 250 Hz) with per-channel spectral and statistical descriptors. Training uses subject-independent splits to avoid leakage, class weighting, and data augmentation (including MixUp); hyperparameters are selected via randomized search with refinement. The model is trained on a publicly available MDD dataset and subsequently applied, without fine-tuning, on an independent acquisition of 20 subjects recorded with a portable three-channel device; we report both window-level and subject-level (majority-vote) performance. On the held-out test subjects from the public dataset, the hybrid model attains 93.43% window-level accuracy. The independent evaluation is reported as a preliminary external feasibility analysis; given the small cohort, we report subject-level performance with 95% confidence intervals to reflect uncertainty and avoid over-interpreting cross-device generalization. The model occupies approximately 40.19 MB on disk, and the architecture is compatible with post-training int8 (TFLite) quantization for resource-constrained hardware. These results, obtained on limited samples, support the feasibility of three-channel EEG for major depressive disorder detection using a lightweight hybrid architecture and motivate prospective clinical validation, on-device inference and quantization studies, and broader evaluation across centers and devices. Full article

Basalt fiber-reinforced concrete is increasingly being used in shotcrete support systems for rock mass excavation engineering due to its superior mechanical properties and durability. Rapid freeze–thaw cycling tests were performed to simulate freeze–thaw conditions in order to meticulously investigate the dynamic and static fracture behaviors of basalt fiber-reinforced concrete in freeze–thaw environments. Then, utilizing a Split Hopkinson Pressure Bar (SHPB) system and rock testing equipment, dynamic and static fracture tests were performed on developed Mode I, mixed-mode I/II, and Mode II platform Brazilian disk specimens. Under freeze–thaw conditions, the dynamic and static fracture propagation velocities of specimens with diverse crack propagation modes were determined. Based on this, LS-DYNA numerical simulations were used to perform inverse evaluations of crack propagation processes in specimens with varied fracture modes and Mode I fracture specimens with variable basalt fiber contents. We were able to calculate the effective stress field distributions during crack propagation with dynamic loading. The data indicate that different fracture modes present significantly distinct crack propagation issues. Pure Mode I fracture specimens exhibit the most straightforward crack propagation, with a maximum effective stress of roughly 25 MPa after crack penetration. With a maximum effective stress of around 31 MPa following crack penetration, the mixed-mode I/II fracture specimens exhibit considerable propagation difficulties. Mode II fracture specimens are the hardest to propagate after crack penetration because of their maximum effective stress of 64 MPa. Additionally, the optimal basalt fiber content was determined to be in the range of 0.35% to 0.45%, at which the concrete exhibited the best fracture toughness and freeze–thaw resistance. Furthermore, the evolution characteristics of the displacement of the crack tip opening under different fracture modes are revealed. A theoretical basis for stability analysis and design of excavation engineering structures under dynamic stress and associated freeze–thaw conditions is provided by the study’s findings. Full article

To investigate the dynamic instability mechanism of surrounding rock in deep, rockburst-prone coal seams, a Split Hopkinson Pressure Bar (SHPB) system was utilized to carry out dynamic impact compression tests on Rock–Coal–Rock (RCR) composites featuring four different seam dip angles, namely 0°, 15°, 30°, and 45°. We systematically analyze incorporating high-speed imaging, the mechanical properties, energy evolution, and progressive failure characteristics of the composites under various strain rates. The results indicate that the dynamic compressive strength and elastic modulus of the composites exhibit a significant strain-rate hardening effect. With the increase in the dip angle of the coal seam, the compressive strength of the specimen decreases accordingly. Specifically, the range of 15–30° is identified as a critical transition zone where the failure mode shifts from matrix-dominated bearing to interfacial slip instability. At an impact pressure of 0.12 MPa, the compressive strength drops by 36.9% within this interval. Furthermore, the energy distribution is profoundly modulated by the geometric characteristics of the interface. As the dip angle increases, the degree of wave impedance mismatch at the coal–rock interface intensifies, leading to a sharp rise in the reflected energy ratio (up to 80.7%) and a pronounced attenuation of transmitted energy. Notably, the dissipation energy per unit volume increases with the dip angle, revealing that interfacial sliding and frictional work become the primary energy dissipation pathways under large-inclination conditions. High-speed camera monitoring confirms that the instability mechanism shifts from axial splitting/tension to an interfacial shear-slip mode as the dip angle increases. These findings provide a scientific reference for the stability evaluation of roadway surrounding rock and the prevention of dynamic disasters. Full article

Background/Objectives: Nutritional management is central to the care of patients with end-stage renal disease (ESRD), yet malnutrition often remains under-recognized due to gaps in nursing knowledge and competencies. This study aimed to develop and validate the Nursing Education and Competencies in Nutrition for Patients with CKD in ESRD (NECN-ESRD) questionnaire, designed to assess nephrology nurses’ competencies, attitudes, and practices in nutritional care. Methods: A methodological and cross-sectional design was adopted, following the COnsensus-based Standards for the selection of health Measurement INstruments (COSMIN) recommendations for instrument development. The process comprised five phases: construct definition and item generation, expert consultation and revision, quantitative content validity analysis, pilot testing, and psychometric testing. Data were collected between August and September 2025 from 405 nephrology nurses across Italy. Exploratory Factor Analyses (EFAs) and Confirmatory Factor Analyses (CFAs) were conducted on split samples (60/40), and key psychometric properties were evaluated. Results: EFA identified a four-factor structure—Recommendations, Attitudes, Practice, and Advanced Competencies—which was confirmed through CFA with good fit indices [Comparative Fit Index (CFI) = 0.995, Tucker–Lewis Index (TLI) = 0.994, Root Mean Square Error of Approximation (RMSEA) = 0.07]. A higher-order model further improved fit (CFI = 0.994, RMSEA = 0.029), explaining 68.2% of variance. Internal consistency was excellent (ω = 0.89–0.96), test–retest reliability showed perfect agreement [Intraclass Correlation Coefficient (ICC) = 1.00], and invariance testing supported equivalence across educational and experience levels. Conclusions: The NECN-ESRD demonstrated strong validity, reliability, and stability, providing a robust and context-specific tool to assess and enhance nurses’ competencies in nutritional care for ESRD patients. Its application can support targeted educational interventions, improve clinical practice, and contribute to enhancing the quality of nutritional care for patients with ESRD within healthcare systems. Full article

Background: Clinicians rely on clinical judgement and vital sign monitoring to identify patient deterioration, commonly supported by systems such as the National Early Warning Score 2 (NEWS2). However, NEWS2 is associated with a high false-positive burden, contributing to alert fatigue in increasingly pressured clinical environments. Consequently, there is a growing need for early warning systems (EWS) that not only detect deterioration but do so with higher precision to prioritise clinically meaningful alerts. We aimed to develop and validate a prognostic EWS capable of predicting real-time clinical deterioration in hospitalised adult patients. Methods: We conducted a retrospective observational cohort study using routinely collected Electronic Patient Record (EPR) data. A Cox proportional hazards model with time-varying covariates was developed to estimate dynamic risk of deterioration. Deterioration was defined as unplanned transfer to intensive care, unplanned surgery, or in-hospital death. Data for model development comprised 37,989 adult inpatient episodes admitted between January 2022 and October 2024, and were initially split into training, temporal validation and test datasets. An extended evaluation period included 11,048 patients admitted through September 2025. Model performance was compared with NEWS2 at the emergency-response threshold (≥7). Results: The final model produced a tiered “traffic-light” risk profile and demonstrated substantially higher precision than NEWS2 while maintaining comparable recall in our test data. At the red alert threshold, precision was 60% compared with 16% for NEWS2 ≥7, with 82% versus 43% of alerts occurring within 24 h of deterioration. Performance remained consistent across the extended evaluation period. Conclusions: A survival-based EWS incorporating time-varying covariates achieved higher precision and improved temporal alignment with deterioration events compared with NEWS2. A tiered amber–red alert framework may support more targeted escalation, reduce alert fatigue, and enhance early identification of clinical deterioration. Full article

Perception-level interference patterns, such as abutting gratings that induce illusory contours and pseudoisochromatic dot camouflage, can trigger failures that are not well captured by conventional corruption benchmarks. We construct an illusion-driven evaluation suite based on MNIST variants, including AG-MNIST and Ishihara-MNIST, under a unified $224\times 224\times 3$ pipeline with fixed train/validation/test splits. Building on a multi-domain empirical risk minimization (ERM) baseline, we present AugMix–DeepAugment Consistency Training (ADCT), a training-time recipe that combines mixed augmentations, DeepAugment-style image distortions, and Jensen–Shannon consistency regularization. Across multiple ImageNet-pretrained backbones and multiple random seeds, ADCT improves robustness on the five-set illusion OOD suite (OOD5) on average while simultaneously improving probabilistic calibration, as measured by ECE together with proper scoring rules (NLL and Brier score). For ResNet-50, ADCT yields a substantial gain on OOD5 relative to the Ishihara-only baseline (S0), increasing accuracy from 29.0% to 59.7% and reducing NLL from 7.44 to 1.11. To assess external validity, we additionally report results on CIFAR-10 and CIFAR-10-C and compare against representative augmentation-based baselines (including PixMix), contextualizing the robustness–calibration trade-off on a widely used natural-image robustness benchmark. These results suggest that consistency-based augmentation recipes can improve both robustness and confidence reliability under structured, illusion-like shifts without changing inference-time architectures. Full article

Background: Periprosthetic joint infection (PJI) after hip arthroplasty for proximal femur fracture is a severe complication, and early postoperative identification remains challenging. This study developed and validated machine learning (ML) models for the early prediction of 90-day EBJIS 2021 “confirmed” PJI using routinely available perioperative data. Methods: We performed a single-center retrospective study including 1182 consecutive adults undergoing primary hip arthroplasty for proximal femur fracture (2015–2022). Forty-seven perioperative candidate predictors were extracted, including early postoperative laboratory values (postoperative day 1–2 and maxima within 72 h). Six algorithms were trained and compared (logistic regression, random forest, support vector machine, multilayer perceptron, XGBoost, and stacking ensemble) using a stratified 80/20 training–test split with 10-fold cross-validation, grid-search hyperparameter tuning, and class weighting. A sensitivity-prioritizing classification threshold was derived using training data only and applied unchanged to evaluation cohorts. Uncertainty was estimated via 1000 bootstrap iterations. Calibration was assessed using the Brier score and calibration intercept/slope. Temporal validation was conducted in a same-center 2023 cohort (n = 147). Model explainability used SHAP. Results: EBJIS-confirmed 90-day PJI occurred in 58/1182 (4.9%) patients. In held-out testing, the final XGBoost model demonstrated good discrimination (AUC 0.889, 95% CI 0.804–0.960) with good overall calibration (Brier score 0.043). Using a prespecified sensitivity-prioritizing threshold selected in the training set, test-set sensitivity was 100%, specificity 58.5%, PPV 11.4%, and NPV 100%. The stacking ensemble yielded the highest discrimination (AUC 0.937; 95% CI 0.89–0.98). In temporal validation (same-center 2023 cohort; n = 147), model performance remained stable (AUC 0.892; sensitivity 85.7%; NPV 99.1% at the prespecified threshold). Calibration was favorable in the development cohort (Brier 0.041; intercept −0.04; slope 0.96) and in 2023 (Brier 0.038; intercept −0.06; slope 0.94). SHAP identified postoperative C-reactive protein, operative duration, body mass index, ASA class, and serum sodium as the most influential predictors. Conclusions: ML models, particularly XGBoost, supported early postoperative risk stratification for 90-day EBJIS-confirmed PJI after fracture-related hip arthroplasty, with a consistently high NPV and stable calibration in a temporally independent same-center cohort. Prospective multi-center validation and impact evaluation are needed before clinical implementation. Full article

To reveal the cumulative damage mechanism of surrounding rock with initial damage under cyclic blasting loads during tunnel reconstruction and expansion, this study combines theoretical modeling, split Hopkinson pressure bar (SHPB) tests, and three-dimensional numerical simulation. First, based on the Z-W-T model framework, a dynamic damage constitutive model capable of uniformly describing the coupling effects of initial damage and dynamic disturbance is constructed by introducing a damage evolution equation based on the Weibull distribution and an initial damage variable D₀. Second, SHPB impact tests are conducted on sandstone specimens with different D₀ values under various strain rates to obtain their dynamic mechanical responses. The model parameters are calibrated and its validity is verified. Finally, the validated model is implemented in ABAQUS via a user material subroutine to establish a 3D finite element model of the tunnel reconstruction and expansion, and a numerical test with seven cyclic blasting events is performed. The results show that the dynamic compressive strength of the surrounding rock increases significantly with increasing strain rate, but D₀ has a clear weakening effect, which is amplified under high strain rates. Numerical simulation reveals that the damage in the surrounding rock accumulates nonlinearly with the number of blasts. The incremental expansion of the damage zone after the first blast is 1.51 m, decreasing to 0.03 m by the seventh blast, indicating a successively diminishing incremental expansion per blast. This reflects the saturation characteristics of damage accumulation and the diminishing driving effect of subsequent blasts due to energy dissipation and compaction within the already-damaged zone. The study provides key theoretical and analytical tools for evaluating the long-term stability of surrounding rock with initial damage under cyclic blasting. Full article

Background: Peripheral nerve injury can happen for a variety of causes. Despite major breakthroughs in microsurgery, nerve repair results are not always sufficient. Methods: Thirty-two Wistar albino rats were split into four groups: primary nerve repair (PNR), PNR with vitamin D3 treatment, nerve crush injury (NCI), and NCI with vitamin D3 treatment. In the PNR + D3 and NCI + D3 groups, 1 mcg/kg of vitamin D3 was given intraperitoneally on days 1, 3, 5, and 7 of the 12-week healing period. Electrophysiological measurements were taken prior to the injury. At 12 weeks after damage, a hot plate test was performed to assess acute pain, and the electrophysiological measurements were repeated. Before the rats were sacrificed, biopsy samples from the right sciatic nerve were collected for histopathological evaluation. Results: Post-healing action potential values were not statistically different between the PNR and PNR + D3 groups; however, they were considerably lower in the NCI + D3 group than in the NCI group. The reaction time in the hot plate test was considerably slower in the D3-treated groups compared to the control groups. Histopathology score was substantially higher in the PNR + D3 group as compared to the PNR group, and lower in the NCI + D3 group as compared to the NCI group. Conclusions: Other than improved myelination, vitamin D3 treatment following primary repair of transected nerves produced no statistically significant improvement. Vitamin D3 treatment caused a negative impact on the crush injury, as assessed by the findings of histopathology and electrophysiological measurements. Overall, the results indicate that the efficacy of vitamin D3 treatment may vary depending on the type of injury. Full article

The research on nonaqueous two-phase systems, i.e., ternary nonaqueous systems with a liquid–liquid phase split induced by a solid component, is discussed. Previous scattered reports are reviewed and summarized. The first systematic studies are described in detail. These included qualitative testing of numerous ternary systems (a solid component and two liquid solvents, significantly different in polarity) to determine whether a liquid–liquid phase split occurred. Some correlations between this occurrence and the Hofmeister series were suggested. The liquid–liquid equilibrium was determined experimentally in a few systems, and the problems encountered during this determination are discussed. Possible applications and further topics of investigation are suggested. Full article

This study evaluates the mechanical behavior and durability of a silty soil stabilized with Portland cement and recycled ground glass powder (GGP). The porosity–cement index (η/C_iv) was applied to predict unconfined compressive strength ($q_u$), splitting tensile strength ($q_t$), and accumulated mass loss (ALM) under wetting–drying cycles. Mixtures were prepared with cement contents of 3%, 6%, and 9%, GGP contents of 5%, 15%, and 30%, and dry unit weights of 13.5, 14.5, and 15.5 kN/m³, and were cured for 7, 28, and 90 days. The experimental program consisted of a large dataset, comprising 486 mechanical tests (unconfined compressive and splitting tensile strength) and 81 durability tests, providing a robust basis for both empirical modeling and machine learning analysis. The results confirmed a strong power-law relationship between η/C_iv and both $q_u$ and $q_t$, achieving high coefficients of determination (R² > 0.98). The strength coefficient (A) increased consistently with curing time and GGP addition, indicating enhanced pozzolanic reactivity and matrix densification. After 90 days, $q_u$ increased by over 250% and $q_t$ by nearly 700%. Durability tests revealed exponential reductions in ALM with higher density and binder content, achieving values below 0.5% for the densest mixtures, which contained 30% GGP. These findings validate the η/C_iv index as an effective predictor of strength and durability in soil–cement–GGP geomaterials, establishing a solid basis for future integration with machine learning models. The implementation of twenty-eight machine learning presets for predicting $q_u$, $q_t$, and ALM demonstrated that the Matern 5/2 Gaussian Process Regression and the trilayered neural network are the most suitable algorithms, achieving R² values higher than 0.987 in both the validation and testing stages. Full article