Search Results (4,024)

Background/Objectives: Recurrent uterine leiomyosarcoma (ULMS) frequently poses a surgical question because systemic options remain limited and recurrence patterns are heterogeneous. We reviewed the published evidence on repeated and extended cytoreductive surgery for recurrent ULMS, focusing on selection criteria, operative boundaries, and the role of multivisceral, thoracic, and peritoneal-directed procedures. Methods: This narrative review synthesizes peer-reviewed literature on surgically managed recurrent or metastatic ULMS, prioritizing contemporary guidelines, retrospective cohorts, pooled analyses, selected systematic reviews when directly relevant to the surgical question, and published illustrative reports. The search covered records available from database inception through 14 May 2026 and used PubMed/MEDLINE, Web of Science Core Collection, Scopus, Google Scholar, selected publisher databases, and citation-linked records. No new patient-level or institution-specific clinical data are presented. Results: The available evidence is entirely retrospective and strongly affected by selection bias, yet it consistently suggests that the best outcomes are observed when complete gross resection is feasible. Across published series, favorable features include isolated or limited recurrence, longer time to relapse, compartmentalized disease, lung-only metastases, and preserved performance status. Contemporary reports also show that repeat surgery may evolve into extensive multivisceral procedures involving bowel resection, upper-abdominal dissection, urinary tract reconstruction, diaphragmatic resection, and thoracic surgery. Peritoneal-directed CRS/HIPEC-type strategies remain supported mainly by small heterogeneous studies and a ULMS-specific systematic review, reinforcing feasibility but not routine use. Published illustrative reports confirm that serial metastasectomies can occasionally support prolonged survival in exceptional patients, but they cannot establish effectiveness. Conclusions: In highly selected patients, repeated and even extensive cytoreductive surgery may remain a rational disease-control strategy for recurrent ULMS. The central unmet need is not proof that surgery can work in exceptional cases, but better criteria to identify who benefits from iterative resection and when escalation to multivisceral or thoracoabdominal surgery is justified. Full article

Background/Objectives: Inheritance classification in rare diseases remains challenging because curated knowledge is incomplete, heterogeneous, and imbalanced across inheritance categories. Disease-level inheritance modeling can support knowledge organization, annotation review, and hypothesis generation in rare-disease resources. This paper introduces RareFusion-Net, a multimodal benchmark framework for disease-level inheritance classification, and evaluates whether integrating ontology-enriched disease text, structured epidemiological metadata, and gene-association information improves prediction in curated rare-disease knowledge bases. RareFusion-Net is intended for knowledge modeling, not individual patient diagnosis. Methods: We developed RareFusionBalanced, a gated multimodal fusion model that combines biomedical disease descriptions, structured metadata, and gene-related information using auxiliary supervision. Ontology-enriched disease text was treated as the dominant semantic modality, while tabular and gene modalities were incorporated as complementary evidence when available. Robustness was improved using balanced regularization, selective transformer fine-tuning, dropout, weight decay, label smoothing, early stopping, and prediction aggregation across random seeds. Evaluation included accuracy, macro-F1, micro-F1, macro-AUC, mean average precision, calibration metrics, class-wise analysis, statistical testing, and ablation experiments. Results: RareFusionBalanced achieved 0.7382 test accuracy, 0.6284 macro-F1, 0.7382 micro-F1, 0.9183 macro-AUC, and 0.6686 mean average precision. Calibration was favorable, with an expected calibration error of 0.0395 and a Brier-OVR of 0.0528. The multimodal model slightly outperformed TextOnly-TransformerBalanced, but improvement over the best TF-IDF baseline was not statistically significant. Ablation showed ontology-enriched text as the strongest modality, with gene associations adding complementary value. Conclusions: RareFusion-Net provides a practical benchmark for ontology-aware rare-disease inheritance modeling. Results suggest selective multimodal benefit while highlighting minority-class difficulty, limited statistical superiority, need for external validation, and improved biological interpretability. Full article

Oxidative stress is a primary driver of sperm deterioration during chilled storage of poultry semen, and identifying effective natural antioxidant supplements for semen extenders is an important practical goal for poultry reproductive management. This study evaluated the protective effects of ginseng extract (Panax ginseng) supplementation on sperm viability, motility, oxidative stress biomarkers, antioxidant defense, and fertility in chilled Leung Hang Kao rooster semen. Pooled semen was diluted in IGGKPh extender supplemented with ginseng extract at 0, 1, 2, 3, or 4 mg/mL and stored at 5 °C for 0, 24, and 48 h. Sperm viability, total motility, progressive motility, malondialdehyde (MDA) concentration, total antioxidant capacity (T-AOC), glutathione peroxidase (GPx) activity, catalase (CAT) activity, and fertility following artificial insemination were evaluated at each time point. All ginseng-supplemented groups showed significantly lower MDA concentrations and higher GPx activity than the unsupplemented control throughout storage. At 48 h, total motility and progressive motility were highest in the 2 and 3 mg/mL groups, while T-AOC was best maintained in the 1 and 2 mg/mL groups. CAT activity did not differ significantly among groups at 48 h (p = 0.2498). Fertility was significantly higher in the 1 and 2 mg/mL groups than in the control after 24 and 48 h of storage, and the alignment between T-AOC and fertility across storage time points indicated that overall antioxidant buffering capacity was a stronger determinant of fertilizing competence than individual enzyme activities or MDA concentration alone. Concentrations of 3–4 mg/mL, despite producing lower MDA at 48 h, did not confer superior fertility outcomes, suggesting a hormetic dose–response relationship. Based on integrated evidence from sperm quality, antioxidant status, and in vivo fertility, ginseng extract supplementation at 1–2 mg/mL is recommended as the most suitable range for preserving chilled Leung Hang Kao rooster semen and may represent a practical natural antioxidant strategy for Thai native poultry breeding programs. Full article

Objective: To analyze the perceived facilitators and barriers promoting and hindering the clinical application of the best evidence of clean intermittent catheterization (CIC) in patients after radical hysterectomy (RH). Methods: This study employed a convergent parallel mixed-methods design. Participants included patients undergoing CIC after RH, medical and nursing practitioners and managers in the gynecological department and outpatient clinics at a tertiary-level women’s and children’s hospital in Chengdu. They were included in both components separately. Interview data were managed using Nvivo 11.0 software and analyzed through directed content analysis. Quantitative data were analyzed using SPSS 29.0 statistical software. Results: A questionnaire survey was conducted among 156 healthcare providers and 300 patients. Qualitative interviews were conducted with 11 healthcare workers and 12 patients. At the evidence itself level, evidence meeting clinical needs and evidence lacking practical applicability, respectively, promoted and hindered clinical implementation of the best evidence. At the potential adopters’ level, healthcare professionals’ insufficient professional competence, low willingness to promote implementation, numerous concerns, and lack of autonomy and awareness regarding the importance of the task were significant barriers, but they maintained an overall positive attitude toward the application. At the practical environment level, patient-related perceived barriers predominantly hindered evidence implementation. Additionally, a supportive practice atmosphere, economic feasibility, and talent development opportunities served as key facilitators. However, existing nursing practice content and workflows directly impacted evidence adoption. Conclusions: The promotion and barriers to the clinical application of the best evidence for CIC in RH postoperative patients are multifaceted. Targeted intervention strategies must be developed to facilitate the effective translation of evidence into clinical practice. Full article

Ceramic additive manufacturing offers strong potential for fabricating geometrically complex and application-specific components, yet achieving reliable dimensional fidelity remains challenging because dimensional deviation is governed by highly coupled material, process, thermal, and environmental factors. To address this problem, this study proposes an uncertainty-aware computational framework for dimensional error prediction in ceramic 3D printing under variable material and process conditions. The contribution is positioned as a system-level integration of established learning, uncertainty estimation, calibration, and reliability-interpretation components within a ceramic additive manufacturing dimensional-error prediction workflow, rather than as a fundamental methodological breakthrough. The validation is conducted using the publicly available Ceramic 3D Printing Process Control Dataset, a 1000-sample tabular dataset, and the resulting findings are therefore interpreted as dataset-specific computational evidence rather than direct proof of industrial deployment readiness. The methodology begins with a structured data-driven preprocessing pipeline that transforms the Ceramic 3D Printing Process Control Dataset into a multi-condition feature space through data cleaning, one-hot material encoding, min–max normalization, and engineered descriptors capturing extrusion–speed balance, thermal gradients, cooling intensity, deposition density, and material-conditioned interactions. A multi-branch deep computational architecture is then developed to encode material, process, thermal-environmental, and engineered-feature streams separately, followed by adaptive cross-condition fusion to learn nonlinear dependencies across ceramic printing regimes. To improve reliability beyond deterministic regression, the framework jointly models aleatoric and epistemic uncertainty and incorporates calibration refinement to align predictive confidence with observed error behavior, thereby enabling preliminary reliability-oriented interpretation of stable and high-risk operating conditions. Experimental results demonstrate that the full model achieves the best overall within-dataset performance, with a test MAE of 0.0118, RMSE of 0.0172, $R^2=0.999$, MAPE of 1.74%, calibration error of 0.003, PICP of 0.996, reliability score of 0.992, and a stable prediction rate of 98.7%. Although these values indicate strong predictive behavior under the current structured dataset, the exceptionally high $R^2$ should be interpreted cautiously because external experimental validation, larger measured datasets, and cross-machine ceramic printing trials are still required. These findings show that the proposed framework provides an effective system-level computational strategy for dataset-specific reliability-aware dimensional quality prediction in ceramic additive manufacturing and offers a preliminary data-driven foundation for uncertainty-aware intelligent process optimization. Full article

Neocinnamomum caudatum (Nees) Merr. is an underutilized woody oil plant with seeds rich in long-chain fatty acids and polyunsaturated fatty acids. Reliable quantitative real-time PCR (RT-qPCR) analysis is essential for investigating the molecular mechanisms underlying seed oil biosynthesis, but suitable reference genes have not yet been validated in this species. Here, seven candidate reference genes, namely EF-1α, ACT2, ACT11, UBQ11, TUA, F-BOX, and GAPDH, were selected from transcriptomic data and evaluated in leaves, flowers, and developing seeds of N. caudatum. Their expression stability was assessed using geNorm, NormFinder, and BestKeeper, followed by comprehensive ranking with RankAggreg. Among all tested samples (leaves, flowers and developing seeds combined), GAPDH was identified as the most stable reference gene, whereas EF-1α was the least stable. For developing seeds alone, TUA showed the highest stability, while EF-1α exhibited poor stability. In leaf and flower samples, ACT11 was the most stable gene, whereas TUA was unsuitable for normalization. The expression patterns of NcFAD2 and NcFatB, two genes involved in fatty acid biosynthesis, were used to validate the selected reference genes. Stable reference genes and the optimized multi-gene combination generated consistent expression profiles, while unstable reference genes caused evident distortion. This study provides the first systematic evaluation of reference genes for qRT-PCR analysis in N. caudatum and offers a practical foundation for future functional studies of lipid metabolism in this woody oil plant. Full article

Sustainability transitions confront firms with the following informational paradox: the regulatory pressure to innovate green has intensified, yet the knowledge required to do so is dispersed across agencies, sectors, and jurisdictions that rarely speak to one another. Treating data as a strategic factor of production, this paper asks whether and how opening public data—the systematic release of government-held datasets—reconfigures the conditions under which firms generate green innovation. We model the green-innovation ecosystem as a Complex Adaptive System (CAS) in which heterogeneous, bounded-rational agents co-evolve with a data-mediated selection environment. Within this frame, public data openness (PDO) is not marginal input but an exogenous shock to the fitness landscape that propagates through three coupling channels—supply–demand alignment, recalibration of government intervention, and amplification of green credit. Formal derivations link each channel to a testable proposition, and a multi-period Difference-in-Differences (DIDs) design built on the staggered roll-out of Chinese municipal open-data platforms identifies the causal effects, with Callaway–Sant’Anna estimators and double/debiased machine learning (DDML) addressing recent econometric critiques. The evidence supports each proposition and reveals the following distinctive heterogeneity signature consistent with absorptive-capacity heterogeneity: the policy is most consequential where agents and ecosystems are best able to convert data into knowledge. Reframing PDO as a systemic enabler clarifies why uniform rollouts yield uneven returns and motivates a tiered design that scales with the absorptive capacity of recipient firms and regions. Full article

This article examines five Qurʾānic lexical items—surādiq (Q 18:29), qiṭṭ (Q 38:16), ḥiṭṭah (Q 2:58; 7:161), fūm (Q 2:61), and yaqṭīn (Q 37:146)—through a theoretical framework that combines multilingualism in Arabia before Islam with muqārana, understood as comparative philological practice. Rather than simply asking whether each word is Arabic or foreign, the article evaluates each case through Qurʾānic context, Arabic morphology and lexicography, phonotactic markedness, comparative Semitic, Iranian, and Mediterranean evidence, variant readings (qirāʾāt), and early exegetical reception (tafsīr). Surādiq illustrates Iranian–Aramaic mediation in eschatological imagery; qiṭṭ and ḥiṭṭah show how documentary and religious-formulaic semantics may preserve older Semitic contact strata; fūm demonstrates how a Qurʾānic food term can be pulled between an archaic Arabic grain/bread meaning, non-canonical reading tradition, and harmonisation via Biblical comparison; and yaqṭīn functions as a control case against the over-identification of borrowings. The article argues that Qurʾānic vocabulary is best studied as multilingual lexical memory: a field in which etymology and exegesis interact without collapsing into a binary opposition between Arabic and foreign vocabulary. Full article

Background/Objectives: Latent fingerprints are routinely recovered from conventional porous and non-porous substrates; however, biologically active surfaces such as plant leaves are generally regarded as unsuitable for dactyloscopic evidence. Because vegetation is frequently present at crime scenes, this study aimed to systematically evaluate whether plant leaves can retain usable friction ridge detail and to determine the durability and forensic value of such traces under laboratory and outdoor conditions. Methods: Latent fingerprints were deposited on leaves of multiple plant species (maple, ash, dandelion, bird cherry, chestnut, climbing ivy, and five-leaved ivy) under dry and hydrated conditions and at defined time intervals after deposition. Visualization was performed using several powders, with SupraNano Fluorescent Green magnetic powder providing the best performance. Developed impressions were photographed using controlled illumination and evaluated using automated quality assessment (NFIQ 2.0) and comparison software (Innovatrics IDkit 9.1.7.1004). Additional experiments examined living, growing leaves exposed to natural weather conditions for extended periods. Results: Usable ridge detail was successfully visualized on all tested species. Bottom leaf surfaces and hydrated samples generally provided better preservation and contrast. Identifiable traces persisted for up to 20 h on detached leaves and for up to 35 days on living leaves despite growth-related deformation. Under outdoor exposure, fingerprints on ivy remained visible and comparable for up to 60 days. Although overall automated quality scores were reduced by background venation, selected impressions achieved measurable comparison scores and successful matches. Conclusions: Plant leaves can serve as unconventional yet viable carriers of latent fingerprints. Magnetic fluorescent powder development combined with careful documentation enables recovery of forensically useful ridge detail even after prolonged environmental exposure. These findings expand the range of substrates that should be considered during crime scene processing and provide practical guidance for evidence collection on vegetation. Full article

This paper summarizes an analysis of the Joint Clinical Assessment (JCA) subgroup’s recommendations for systematic literature reviews (SLRs). While the JCA offers clear guidance on study classification, exclusion criteria reporting, and PRISMA diagram use, several of its recommendations diverge from established best practices in evidence-based medicine (EBM). A comparison with recognized guidelines, such as those from Cochrane and EUnetHTA, reveals that the JCA guidance may lack reliability, comprehensiveness, and reproducibility. Aligning JCA recommendations with gold standards in SLR methodology would address these shortcomings and enhance methodological rigor. Full article

Political violence poses a significant challenge to global stability, underscoring the need for comparative analytical models that support analytical interpretation of structured conflict data. This paper presents a comparative evaluation of 12 machine learning approaches, including traditional supervised models, deep learning architectures, and zero-shot large language models, for the classification of political violence events using the Armed Conflict Location and Event Data Project (ACLED) dataset (2010–2020, over 40,000 events). The results demonstrate that, on short structured event text represented via TF-IDF, fine-tuned traditional machine learning models achieve stronger performance than zero-shot LLM approaches and deep learning models on structured event data. We further introduce a multilingual classification framework for English and Urdu news content, illustrating cross-lingual transfer robustness using machine-translated Urdu data; results reflect translation-based evaluation conditions and should not be interpreted as performance on naturally occurring low-resource Urdu political-event text. As an exploratory extension, the framework is applied to 57,700 tweets related to the Article 370 crisis in Kashmir to illustrate applicability to unstructured social media text; given that the best Twitter model (55% accuracy) falls below the 69% majority-class baseline, these results should be interpreted solely as coarse discourse indicators and not as a validated classification component. Unlike prior work, this study systematically combines multilingual evaluation with zero-shot LLM analysis for political event classification. Geographic out-of-sample validation (leave-one-country-out or leave-one-region-out) was not conducted; the reported performance should therefore not be interpreted as evidence of cross-regional generalizability without further experimentation. The findings highlight practical considerations for designing data-driven analytical frameworks for conflict monitoring and analytical decision support. Full article

Background/Objectives: Artificial intelligence (AI) offers a promising approach for detecting and classifying symptom severity in patients with Parkinson’s disease (PD). The objective was to provide an overview of AI methods performance used for this classification through a systematic review and meta-analysis conducted in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Methods: The Google Scholar, IEEE Xplore, PubMed/MedLine, and ScienceDirect databases were searched for the period 2015–2025. The studies included were original, peer-reviewed studies written in English that addressed an AI method based on machine learning (ML) or deep learning (DL) for the classification of PD patients. The dataset used had to be “Gait in Parkinson’s Disease,” in which the severity of disease symptoms was assessed using the Hoehn and Yahr (H&Y) scale. Studies had to report at least one of the five performance metrics: accuracy, sensitivity, specificity, precision, and F1 score. Two reviewers independently selected articles, assessed the risk of bias using PROBAST (Prediction Model Study Risk of Bias Assessment Tool), and extracted data. The logit-transformed values were pooled separately by performance metrics and by severity level using a random-effects model. Cochran’s Q test, the I² statistic, and inter-study variability (τ²), computed using the generalized inverse variance method with the restricted maximum likelihood model, were used to assess heterogeneity. Forest plots with 95% confidence intervals were used to present the results. Possible causes of heterogeneity were explored using a subgroup analysis (ML vs. DL) and a sensitivity analysis. Finally, publication bias (Egger’s test) and the certainty of the evidence (using GRADE—Grading of Recommendations Assessment, Development, and Evaluation) were assessed to verify the generalizability of the results. Results: Among the 257 unique records, 12 studies were included. The methods demonstrated very high overall performance (>92%): accuracy (96.4%, 95% CI: 95.9–96.9%), specificity (97.7%, 95% CI: 97.3–98.1%), sensitivity (94.0%, 95% CI: 92.7–95.2%), precision (93.4%, 95% CI: 92.0–94.6%), F1 score (92.1%, 95% CI: 90.6–93.4%). Accuracy, specificity, and precision were high for all H&Y levels. However, the more advanced the symptoms, the lower the sensitivity (97.3% for H&Y0 vs. 92.1% for H&Y3). ML models achieved the best results for classifying healthy patients (H&Y0: 95.7% to 98.2%), while DL approaches performed better for classifying higher severity levels (>92%). Heterogeneity and inter-study variability were moderate (I²: 40–50% and τ²: 0.3–0.4) for precision and F1 score, and high (I² > 90% and τ² > 0.6) for accuracy, specificity, and sensitivity. The GRADE analysis revealed low-quality evidence for precision and F1 score and very-low quality for accuracy, specificity, and sensitivity. Conclusions: Thus, AI-based wearable gait assessment devices show great promise in terms of aiding clinical decision-making and treatment personalization. However, further research using a rigorous methodology (PROBAST) is needed to ensure the generalizability of the results and the clinical viability of the proposed solutions. Full article

Seasonal freeze–thaw cycling progressively rearranges pores and propagates microcracks in silty clay, reducing the reliability of cold-region earthworks. This study evaluated a bio–polymer stabilization strategy combining microbially induced carbonate precipitation (MICP) with anionic polyacrylamide (APAM) to improve mechanical performance and freeze–thaw durability. Six groups were prepared at identical moisture and compaction conditions: water, APAM, and four MICP–APAM groups with bacterial optical densities (OD600) of 0.8, 1.0, 1.2, and 1.4. Unconfined compressive strength, unconsolidated-undrained triaxial compression, ultrasonic pulse velocity, and SEM, TG/DTG, XRD, and FTIR analyses were conducted before and after freeze–thaw cycling. The M1.0-APAM group showed the best overall performance, with UCS values of 1.35 MPa before cycling and 0.89 MPa after nine cycles, together with high shear resistance and ultrasonic velocity. Lower bacterial concentration provided insufficient cementation, whereas higher concentrations promoted non-uniform carbonate deposition, pore heterogeneity, and local stress concentration. Microstructural evidence indicated that OD600 ≈ 1.0 produced a relatively homogeneous network of fine carbonate clusters and polymer-associated films, with calcite formation supported by TG/DTG and XRD. The results show that MICP–APAM treatment enhances silty clay primarily through coordinated mineralization uniformity, pore refinement, and polymer bridging, providing a sustainable stabilization option for seasonally frozen soils. Full article

Road infrastructure accounts for a substantial and systematically under-reported fraction of construction-related embodied carbon globally. Despite rapid network expansion across sub-Saharan Africa, no peer-reviewed study identified in the databases searched has established a quantified embodied-carbon baseline for Ghanaian road construction, creating a notable gap in national carbon accounting and low-carbon procurement policy. This study addresses that gap through two integrated components: a PRISMA 2020-guided systematic review of road-LCA and embodied-carbon literature, and a first-pass scenario model for Ghanaian low-volume paved roads (LVRs) bounded at A1–A3 (cradle-to-gate). Database searches of Scopus and Web of Science (14 March 2026) returned 3193 records; following deduplication and two-stage screening, 574 studies were included in the review. A staged harmonisation procedure converted 211 benchmark-shortlisted studies to comparable units, yielding a harmonisation subset of 29 studies and a final benchmark pool of 10 studies expressed as kgCO₂e per lane-kilometre (3.5 m lane width). The scenario model applies emission factors from the ICE Database (Educational V4.1, 2025) to three pavement configurations drawn from the Ghana Manual for Low Volume Roads (Parts B and D), all surfaced with double bituminous surface treatment (DBST); Otta seal is evaluated as a sensitivity case. Results show A1–A3 embodied carbon of 14,165 kgCO₂e/lane-km for Scenarios S1 and S3 (SC2/TLC 0.01 and SC4/TLC 1.0, respectively) and 12,564 kgCO₂e/lane-km for Scenario S2 (SC3/TLC 0.3). Bituminous binder accounts for 30–34% of A1–A3 emissions despite representing less than 1% of pavement mass, identifying binder supply as the primary carbon lever. The two most structurally comparable benchmark studies, chip-seal treatments in the USA, bracket the Ghana values at 12,687–16,400 kgCO₂e/lane-km, providing external plausibility validation. To the best of our knowledge, this study delivers a peer-reviewed, reproducible A1–A3 (cradle-to-gate) carbon baseline for Ghanaian LVR construction, a PRISMA-compliant synthesis of road embodied-carbon evidence, and a documented framework for early-stage carbon benchmarking in West African road infrastructure planning. Full article

Reliable diagnosis of incipient transformer faults is essential for preventing catastrophic failures and enabling predictive asset management in power systems. Although dissolved gas analysis (DGA) is the most established diagnostic tool for assessing transformer internal condition, fault discrimination remains difficult when gas features are highly correlated, redundant, and only partially separable across fault classes. This study presents a PCA-enhanced artificial neural network (ANN) framework for multiclass transformer fault diagnosis using DGA data. The method is developed on 595 samples classified into six IEC 60599 fault categories and uses a 15-feature representation comprising raw gas concentrations, total hydrocarbon content, and engineered gas-ratio descriptors. To identify an evidence-based diagnostic representation, principal component analysis (PCA) was evaluated across all dimensionalities from k = 1 to 15 before ANN training. The proposed model was benchmarked against alternative feature sets and conventional classifiers, including Gaussian Naïve Bayes, k-nearest neighbours, support vector machines, and ANN without PCA. The best-performing configuration was obtained at k = 13, yielding a test accuracy of 68.1%, compared with 63.9% for ANN without PCA, 56.3% for raw-gas-only ANN, and 33.6% for the IEC three-ratio feature configuration. In addition to improving diagnostic performance, the PCA stage revealed interpretable component structures associated with dominant gas and ratio patterns underlying fault separation. The results indicate that PCA-based feature extraction improves ANN generalization by reducing redundancy and multicollinearity in DGA-derived variables, and provides a practical, lightweight, and interpretable framework for transformer fault diagnosis. Full article