Search Results (78)

Background: Chest X-ray (CXR) is widely used for the assessment of thoracic diseases, yet automated multi-label interpretation remains challenging due to subtle visual patterns, overlapping anatomical structures, and frequent co-occurrence of abnormalities. While recent deep learning models have shown strong performance, limitations in interpretability, anatomical awareness, and robustness continue to hinder their clinical adoption. Methods: The proposed framework employs a hybrid ConvNeXtV2–Vision Transformer (ViT) architecture that combines convolutional feature extraction for capturing fine-grained local patterns with transformer-based global reasoning to model long-range contextual dependencies. The model is trained exclusively using image-level annotations. In addition to classification, three complementary post hoc components are integrated to enhance model trust and interpretability. A segmentation-aware Gradient-weighted class activation mapping (Grad-CAM) module leverages CheXmask lung and heart segmentations to highlight anatomically relevant regions and quantify predictive evidence inside and outside the lungs. An ontology-driven neuro-symbolic reasoning layer translates Grad-CAM activations into structured, rule-based explanations aligned with clinical concepts such as “basal effusion” and “enlarged cardiac silhouette”. Furthermore, a lightweight out-of-distribution (OOD) detection module based on confidence scores, energy scores, and Mahalanobis distance scores is employed to identify inputs that deviate from the training distribution. Results: On the VinBigData test set, the model achieved a macro-AUROC of 0.9525 and a Micro AUROC of 0.9777 when trained solely with image-level annotations. External evaluation further demonstrated strong generalisation, yielding macro-AUROC scores of 0.9106 on NIH ChestXray14 and 0.8487 on CheXpert (frontal views). Both Grad-CAM visualisations and ontology-based reasoning remained coherent on unseen data, while the OOD module successfully flagged non-thoracic images. Conclusions: Overall, the proposed approach demonstrates that hybrid convolutional neural network (CNN)–vision transformer (ViT) architectures, combined with anatomy-aware explainability and symbolic reasoning, can support automated chest X-ray diagnosis in a manner that is accurate, transparent, and safety-aware. Full article

Legal judgment prediction (LJP) increasingly relies on large language models whose full fine-tuning is memory-intensive and susceptible to catastrophic forgetting. We present LawLLM-DS, a two-stage Low-Rank Adaptation (LoRA) framework that first performs legal knowledge pre-tuning with an aggressive learning rate and subsequently refines judgment relations with conservative updates, using dedicated LoRA adapters, 4-bit quantization, and targeted modification of seven Transformer projection matrices to keep only 0.21% of parameters trainable. From a structural perspective, the twenty annotated legal elements form a symmetric label co-occurrence graph that exhibits both cluster-level regularities and asymmetric sparsity patterns, and LawLLM-DS implicitly captures these graph-informed dependencies while remaining compatible with downstream GNN-based representations. Experiments on 5096 manually annotated divorce cases show that LawLLM-DS lifts macro F1 to 0.8893 and achieves an accuracy of 0.8786, outperforming single-stage LoRA and BERT baselines under the same data regime. Ablation studies further verify the contributions of stage-wise learning rates, adapter placement, and low-rank settings. These findings demonstrate that curriculum-style, parameter-efficient adaptation provides a practical path toward lightweight yet structure-aware LJP systems for judicial decision support. Full article

As a non-invasive diagnostic technique, evolved gas analysis (EGA) holds significant value in assessing the insulation conditions of critical equipment such as power cables. Current analytical methods face two major challenges: insulation materials may undergo multiple aging mechanisms simultaneously, leading to interfering characteristic gases; and traditional approaches lack the multi-label recognition capability to address concurrent fault patterns when processing mixed-gas data. These limitations hinder the accuracy and comprehensiveness of insulation condition assessment, underscoring the urgent need for intelligent analytical methods. This study proposes a deep convolutional neural network (DCNN)-based multi-label classification framework to accurately identify the gas generation characteristics of five typical power cable insulation materials—ethylene propylene diene monomer (EPDM), ethylene-vinyl acetate copolymer (EVA), silicone rubber (SR), polyamide (PA), and cross-linked polyethylene (XLPE)—under fault conditions. The method leverages concentration data of six characteristic gases (CO₂, C₂H₄, C₂H₆, CH₄, CO, and H₂), integrating modern data analysis and deep learning techniques, including logarithmic transformation, Z-score normalization, multi-scale convolution, residual connections, channel attention mechanisms, and weighted binary cross-entropy loss functions, to enable simultaneous prediction of multiple degradation states or concurrent fault pattern combinations. By constructing a gas dataset covering diverse materials and operating conditions and conducting comparative experiments to validate the proposed DCNN model’s performance, the results demonstrate that the model can effectively learn material-specific gas generation patterns and accurately identify complex label co-occurrence scenarios. This approach provides technical support for improving the accuracy of insulation condition assessment in power cable equipment. Full article

Background/Objectives: Comprehensive genomic profiling (CGP) is a cornerstone of personalized oncology. However, large-scale, systematic data on the somatic mutation spectrum in Russian cancer patients are scarce. This study aimed to characterize the genomic landscape and assess the potential for matched therapy in a Russian cohort of patients with solid tumors. Methods: This retrospective study included 204 patients with various solid tumors. CGP was performed using the FoundationOne^®CDx (FFPE tissue) and FoundationOne^®Liquid CDx (cfDNA) platforms. The analysis assessed single-nucleotide variants, indels, copy number alterations, gene fusions, tumor mutational burden (TMB), microsatellite instability (MSI), and PD-L1 expression. Results: The most frequently mutated genes were TP53 (61.5%) and KRAS. The median TMB was 4.0 mut/Mb and was significantly lower in stage IV tumors. Significant co-occurrence was observed between KRAS and TP53 mutations, as well as between APC and KRAS mutations, which were particularly characteristic of colorectal cancer. KRAS mutations were associated with higher combined positive score (CPS) values in cases with lung cancer. Based on the CGP results, 44% of patients had findings that supported the use of an approved matched targeted therapy or immunotherapy for their tumor type. An additional 36% of patients had alterations indicating potential benefit from off-label targeted therapy. Conclusions: This study reveals the distinct genomic characteristics of solid tumors in a Russian cohort and confirms the high clinical utility of CGP for identifying actionable targets. Implementing CGP early in the diagnostic process is a necessary step towards realizing personalized treatment strategies for cancer patients. Full article

Ensemble clustering has become a widely used technique for improving robustness and accuracy by combining multiple clustering results. However, traditional ensemble clustering methods often fail to provide fair treatment between groups defined by sensitive attributes. Central to many ensemble methods is the symmetric co-association matrix, which captures pairwise similarity between data points based on their co-occurrence across base clusterings. This paper introduces a fair ensemble clustering method based on the symmetric co-association matrix. The proposed method integrates fairness constraints into the objective function of the ensemble process, using the results from base clusterings that lack fairness considerations as input. The optimization is performed iteratively, and the final clustering results are represented directly by a label matrix obtained efficiently using a coordinate descent approach. By integrating fairness into the clustering process, the method avoids the need for post-processing to achieve fair results. Comprehensive experiments on both real-world and synthetic datasets validate the effectiveness and practicality of the proposed method. Full article

Probiotics are widely used as feed additives in livestock production, yet the overall safety of commercially available veterinary probiotics remains insufficiently assessed. In this study, 33 probiotic products marketed in Northern China were systematically evaluated with respect to strain composition, label accuracy, antimicrobial resistance, and the diversity of resistance genes. A total of 32 Bacillus spp. were isolated, many of which showed resistance to multiple antibiotics. Labeling inaccuracies were prevalent: none of the products specified strain names and numbers, 33% (11/33) failed to report viable bacterial counts, 9% (3/33) lacked their claimed key ingredients, and 21% (7/33) contained isolated strains that did not match the label. High-throughput quantitative PCR (HT-qPCR) analysis further revealed that all 27 tested products harbored abundant antibiotic resistance genes (ARGs), with 241 ARGs and seven mobile genetic elements (MGEs) detected. The ARGs were primarily associated with tetracycline, aminoglycosides, β-lactams, and macrolide–lincosamide–streptomycin B (MLSB) antibiotics, and co-occurrence analysis showed a strong positive correlation between ARG and MGE abundance, with Clostridium and Enterococcus identified as potential hosts. These findings underscore significant quality and safety deficiencies in veterinary probiotics and highlight potential risks to animal, human, and environmental health, emphasizing the relevance of a One Health perspective in probiotic evaluation and regulation. Full article

Bimodal emotion recognition based on audio and text is widely adopted in video-constrained real-world applications such as call centers and voice assistants. However, existing systems suffer from limited cross-domain generalization and monolingual bias. To address these limitations, a cross-lingual bimodal emotion recognition method is proposed, integrating Mamba-based temporal encoders for audio (Wav2Vec2.0) and text (Jina-v3) with a Transformer-based cross-modal fusion architecture (BiFormer). Three corpus-adaptive augmentation strategies are introduced: (1) Stacked Data Sampling, in which short utterances are concatenated to stabilize sequence length; (2) Label Smoothing Generation based on Large Language Model, where the Qwen3-4B model is prompted to detect subtle emotional cues missed by annotators, producing soft labels that reflect latent emotional co-occurrences; and (3) Text-to-Utterance Generation, in which emotionally labeled utterances are generated by ChatGPT-5 and synthesized into speech using the DIA-TTS model, enabling controlled creation of affective audio–text pairs without human annotation. BiFormer is trained jointly on the English Multimodal EmotionLines Dataset and the Russian Emotional Speech Dialogs corpus, enabling cross-lingual transfer without parallel data. Experimental results show that the optimal data augmentation strategy is corpus-dependent: Stacked Data Sampling achieves the best performance on short, noisy English utterances, while Label Smoothing Generation based on Large Language Model better captures nuanced emotional expressions in longer Russian utterances. Text-to-Utterance Generation does not yield a measurable gain due to current limitations in expressive speech synthesis. When combined, the two best performing strategies produce complementary improvements, establishing new state-of-the-art performance in both monolingual and cross-lingual settings. Full article

Background/Objectives: Research on restaurants has traditionally emphasized business operations. Considering restaurants’ growing role in shaping dietary patterns and public health outcomes, this study aimed to map the scope, trends, and gaps in scholarly research addressing food-related aspects of restaurants, excluding business-oriented topics. Methods: A bibliometric analysis was conducted using the Web of Science and Scopus databases. Search terms encompassed multiple restaurant categories, including fast food, fast casual, casual dining, and fine dining. After screening, 956 peer-reviewed English-language journal articles were included. Descriptive performance metrics were calculated, and keyword co-occurrence analysis was conducted. Results: Findings revealed that nutrition-related studies dominate the literature, particularly research linking fast food consumption to obesity and the impact of menu labeling policies on consumer behavior. Food science research was comparatively limited and concentrated primarily on food safety and uses for degraded frying oil. The analysis also highlighted a strong research focus on fast food, while fast casual and fine dining restaurants were notably underrepresented. Conclusions: Future studies should move beyond short-term, cross-sectional designs and incorporate longitudinal approaches to better capture how policy interventions, such as menu labeling and reformulation incentives affect consumer food choices and restaurant offerings over time. Understanding how to reduce restaurants’ contribution to the incidence of diet-related noncommunicable disease risk factors such as obesity and hypertension will require research trials that jointly manipulate key factors such as economic (prices and incentives), structural (recipes, assortment, and operations), and behavioral (choice architecture). Research could also investigate strategies to reduce allergen risks by evaluating standardized training programs and integrated menu/POS disclosure systems. In addition, examination of consumer acceptance of sustainable ingredient substitutions and packaging methods is needed. Full article

We present a bibliometric analysis of unmanned aerial vehicle (UAV) research that replaces simple keyword filtering with a context-aware, two-tier pipeline. Records from Web of Science and Scopus (198,152 total) were harmonized and de-duplicated in three stages (DOI, normalized title, blockwise Jaro–Winkler), yielding 129,124 unique items. To separate UAV work from entomology using overlapping vocabulary (e.g., swarm), we first applied rule-based weak labels with explicit UAV and insect regex families and a UAV context rule for “swarm,” then trained an elastic-net logistic regression on TF–IDF features and tuned the decision threshold to meet a high-precision target on a held-out split. The final corpus comprises 129,099 UAV records. Beyond lexical inventories, a keyword co-occurrence timeline shows reinforcement learning increasingly aligned with path planning and collision avoidance, while constraints such as energy and communication persist. A co-authorship network reveals bridging authors that connect guidance/control, perception, and communication subfields. The results show how UAV research is organized around central scientific problems and identify persistent obstacles such as energy efficiency, communication reliability, and robust decision-making in dynamic conditions. Full article

Modern semantic segmentation models require extensive pixel-level annotations, creating a significant barrier to practical deployment as labeling a single image can take hours of human effort. Active learning offers a promising way to reduce annotation costs through intelligent sample selection. However, existing methods rely on poorly calibrated confidence estimates, making uncertainty quantification unreliable. We introduce Conformal Risk Controlled Active Learning (CRC-AL), a novel framework that provides statistical guarantees on uncertainty quantification for semantic segmentation, in contrast to heuristic approaches. CRC-AL calibrates class-specific thresholds via conformal risk control, transforming softmax outputs into multi-class prediction sets with formal guarantees. From these sets, our approach derives complementary uncertainty representations: risk maps highlighting uncertain regions and class co-occurrence embeddings capturing semantic confusions. A physics-inspired selection algorithm leverages these representations with a barycenter-based distance metric that balances uncertainty and diversity. Experiments on Cityscapes and PascalVOC2012 show CRC-AL consistently outperforms baseline methods, achieving 95% of fully supervised performance with only 30% of labeled data, making semantic segmentation more practical under limited annotation budgets. Full article

The proliferation of image sharing on social media poses significant privacy risks. Although some previous works have proposed to detect privacy attributes in image sharing, they suffer from the following shortcomings: (1) reliance only on legacy architectures, (2) failure to model the label correlations (i.e., semantic dependencies and co-occurrence patterns among privacy attributes) between privacy attributes, and (3) adoption of static, one-size-fits-all user preference models. To address these, we propose a comprehensive framework for visual privacy protection. First, we establish a new state-of-the-art (SOTA) architecture using modern vision backbones. Second, we introduce Graph Convolutional Networks (GCN) as a classifier head to counter the failure to model label correlations. Third, to replace static user models, we design a dynamic personalization module using Federated Learning (FL) for privacy preservation and Reinforcement Learning (RL) to continuously adapt to individual user preferences. Experiments on the VISPR dataset demonstrate that our approach can outperform the previous work by a substantial margin of 6% in mAP (52.88% vs. 46.88%) and improve the Overall F1-score by 10% (0.770 vs. 0.700). This provides more meaningful and personalized privacy recommendations, setting a new standard for user-centric privacy protection systems. Full article

Soil aggregates are crucial for fertility and organic carbon (C) sequestration, with straw decomposition by soil microbes playing a key role in this process. However, the mechanisms of how fertilization and microbes control straw decomposition and the subsequent formation of straw-derived C in soil aggregates are still unclear. Therefore, topsoil samples (0~20 cm) were collected from three fertilization treatments in a long-term (29-year) Mollisol field experiment: (i) no fertilization control, CK; (ii) inorganic fertilizer, IF; and (iii) inorganic fertilizer plus manure, IFM. Thereafter, an in situ micro-plot incubation experiment was conducted without/with ¹³C-labeled straw (abbreviated as CKS, IFS, and IFMS, respectively). Soil aggregates were separated into macro- (>0.25 mm) and microaggregates (<0.25 mm). The aggregate-based changes in straw-derived C content, microbial community composition, co-occurrence network, keystone taxa, and functional characteristics were measured on the 1st, 60th, and 150th day after straw addition. The results showed that straw-derived C content increased averagely by 7 (CKS), 13 (IFS), and 20 times (IFMS) from day 1 to day 150 in the macroaggregates. The straw-derived C content in the microaggregates was the highest in the IFS (0.70%) and IFMS (0.67%) treatments on day 60. After straw addition, the relative abundance of Humicola within the soil macroaggregates significantly decreased from 2.9% (CK) to 1.4% (CKS), and that of Penicillium within the soil microaggregates decreased from 7.5% (IF) to 4.0% (IFS) on day 150. Network analysis revealed greater microbial complexity in microaggregates than in macroaggregates, with fungal keystone taxa responding more strongly to straw than bacterial keystone taxa. The SEM model identified bacterial composition and fertilization as key drivers of straw-derived C formation in macro- and microaggregates, respectively. These findings highlight the distinct roles of bacteria and fungi in various sizes of aggregate and the importance of customized soil management for improving soil fertility and C storage. Full article

In response to China’s dual-carbon strategy, this study proposes a comprehensive analytical framework to identify the evolutionary pathways of key policy tasks in developing a new-type power system. A dual-channel data acquisition process was designed to extract, standardize, and segment policy documents and online texts into a unified corpus. A multi-label BERT classification model was then developed, incorporating domain-specific terminology injection, label-wise attention, dynamic threshold scanning, and imbalance-aware weighting. The model was trained and validated on 200 energy news articles, 100 official policy releases, and 10 strategic planning documents. By the 10th epoch, it achieved convergence with a Macro-F1 of 0.831, Micro-F1 of 0.849, and Samples-F1 of 0.855. Ablation studies confirmed the significant performance gain over simplified configurations. Structural label analysis showed “Build system-friendly new energy power stations” was the most frequent label (107 in plans, 80 in news, 24 in policies) and had the highest co-occurrence (81 times) with “Optimize and strengthen the main grid framework.” The label co-occurrence network revealed multi-layered couplings across generation, transmission, and storage. The Priority Evaluation Index (PEI) further identified “Build shared energy storage power stations” as a structurally central task (centrality = 0.71) despite its lower frequency, highlighting its latent strategic importance. Within the domain of national-level public policy and planning documents, the proposed framework shows reliable and reusable performance. Generalization to sub-national and project-level corpora is left for future work, where we will extend the corpus and reassess robustness without altering the core methodology. Full article

Soil organic carbon (SOC) represents a crucial factor in agricultural production, and its accumulation is influenced by soil microbial community and microbial metabolism. Straw returning combined with decomposing agents is recognized practice to enhance SOC. On the other hand, the impacts of controlled-release nitrogen fertilizer (CR) on the function of the decomposing agent in degrading straw are underexplored. In this study, an incubation experiment with ¹³C labeled straw in three nitrogen fertilizer treatments (CK, no nitrogen applied; UR, urea applied; CR, controlled-release fertilizer applied) was carried out to elucidate how CR regulates the straw decomposition agent and bacterial community to influence the SOC sequestration, based on field experiments. And we examined the changes in soil organic carbon and the stability of the bacterial networks by combining co-occurrence networks and a structural equation model. In the incubation experiment, the results demonstrated that CR increased the relative abundance of straw decomposition agent and straw-derived SOC (SO¹³C). Additionally, CR enhanced the stability of soil bacterial networks, compared with UR, by strengthening the interactions within the soil bacterial community. Pearson correlations confirmed that straw decomposition agent was positively associated with SO¹³C. Moreover, the straw decomposition agent was positively correlated with the activities of the nitrogen-cycling enzyme (urease, N-acetyl-β-glucosaminidase) and carbon-degrading enzyme (β-1,4-glucosidase, cellulase). Furthermore, structural equation modeling indicated that soil inorganic nitrogen played the most direct role in changes in the straw decomposition agent and then indirectly stimulated the activity of cellulase, ultimately increasing straw-derived carbon in the soil. This study elaborates the mechanism of straw returning combined with straw decomposition agent and controlled-release fertilizers to enhance the SOC of coastal saline–alkali soil from the perspective of underground biology. Collectively, the results of this research might improve the management of straw returning and sustainable utilization of fertility in saline–alkali soil. It provides a new perspective on fertilization for increasing soil carbon sequestration in future farmland ecosystems. Full article

A multimodal interactive system, HortiVQA-PP, is proposed for horticultural scenarios, with the aim of achieving precise identification of pests and their natural predators, modeling ecological co-occurrence relationships, and providing intelligent question-answering services tailored to agricultural users. The system integrates three core modules: semantic segmentation, pest–predator co-occurrence detection, and knowledge-enhanced visual question answering. A multimodal dataset comprising 30 pest categories and 10 predator categories has been constructed, encompassing annotated images and corresponding question–answer pairs. In the semantic segmentation task, HortiVQA-PP outperformed existing models across all five evaluation metrics, achieving a precision of 89.6%, recall of 85.2%, F1-score of 87.3%, $mAP@50$ of 82.4%, and IoU of 75.1%, representing an average improvement of approximately 4.1% over the Segment Anything model. For the pest–predator co-occurrence matching task, the model attained a multi-label accuracy of 83.5%, a reduced Hamming Loss of 0.063, and a macro-F1 score of 79.4%, significantly surpassing methods such as ASL and ML-GCN, thereby demonstrating robust structural modeling capability. In the visual question answering task, the incorporation of a horticulture-specific knowledge graph enhanced the model’s reasoning ability. The system achieved 48.7% in BLEU-4, 54.8% in ROUGE-L, 43.3% in METEOR, 36.9% in exact match (EM), and a GPT expert score of 4.5, outperforming mainstream models including BLIP-2, Flamingo, and MiniGPT-4 across all metrics. Experimental results indicate that HortiVQA-PP exhibits strong recognition and interaction capabilities in complex pest scenarios, offering a high-precision, interpretable, and widely applicable artificial intelligence solution for digital horticulture. Full article