Search Results (2,139)

In recent years, deep learning-based vulnerability detection has drawn wide attention for its data-driven ability to analyze code semantics and learn vulnerability patterns without predefined models. However, data distribution differences across projects limit model generalization. Transfer learning provides a solution, yet most studies ignore expert-designed metrics. This paper proposes Decpvd, a data-driven cross-project software vulnerability detection method based on a dual-feature stacking ensemble. It builds an adaptive and transferable model using only code and vulnerability label data from source and target projects. It extracts code semantic features via Gated Graph Neural Networks, incorporates expert metrics from tools, performs cross-domain data-driven modeling with TrAdaBoost, and adaptively fuses the two features through stacking, overcoming fixed-weight fusion limitations. Experiments on six cross-project groups from three real datasets (FFmpeg, LibTIFF, LibPNG) show that Decpvd achieves an average AUC of 0.814, significantly outperforming mainstream baselines. Full article

Background/Objectives: Dysregulation of purinergic signaling, particularly CD73 overexpression, influences tumor progression, immune evasion, and chemoresistance in hepatocellular carcinoma (HCC). We aimed to characterize the transcriptional landscape of this system, identify prognostic markers, and investigate how the tumor microenvironment modulates pharmacological response to combined sorafenib and doxazosin in 3D spheroid models. Methods: We integrated RNA-seq data from The Cancer Genome Atlas—Liver Hepatocellular Carcinoma (TCGA-LIHC) to identify differentially expressed genes, pathway enrichment, gene co-expression networks, prognostic associations, and machine learning-based biomarker selection. Modulation of key targets was assessed in HepG2 and HepG2/LX-2 spheroids treated with sorafenib and doxazosin using qPCR and flow cytometry. Results: Transcriptomics revealed dysregulation and network fragmentation. Specifically, analysis of the TCGA cohort indicated that high expression of ADA, NT5E, and ADORA1 correlated with poor overall survival. Given the critical role of CD73 in therapy resistance, we evaluated these findings in 3D models. Co-treatment significantly downregulated NT5E and ADORA1 mRNA expression, while ADORA2A was specifically reduced in the co-culture setting. For the ADA, effect-size analysis revealed a large magnitude of inhibition in HepG2 spheroids. Although flow cytometry showed that high CD73 protein expression remained stable across treatments in co-culture, the combination therapy overcame stromal protection, significantly increasing apoptosis (active caspase-3) in both mono- and co-culture spheroids compared with vehicle and monotherapy. Conclusions: We identified a purinergic prognostic signature in HCC and demonstrated that the combination therapy of sorafenib and doxazosin targets the adenosine pathway and specific receptors. We show that the stromal microenvironment sustains CD73 protein expression even under transcriptional inhibition, highlighting the critical role of 3D co-culture models in deciphering therapeutic resistance mechanisms. Full article

Background: Pediatric non-Hodgkin lymphoma (NHL) is a heterogeneous malignancy with variable outcomes. Adenosine deaminase (ADA), a key enzyme in purine metabolism, has been implicated in tumor progression and immune evasion, yet its role in pediatric NHL remains underexplored. Methods: This retrospective study included 215 pediatric NHL patients categorized into precursor cell lymphoma (n = 88) and mature cell lymphoma (n = 127) groups based on pathology. Patients were further defined into good (n = 143) and poor prognosis (n = 72) groups according to international response criteria. Serum ADA and other laboratory parameters were measured at diagnosis. Results: Precursor cell lymphomas showed higher rates of bone marrow involvement, peripheral blood involvement, and clinical stage IV disease compared to mature cell lymphomas. ADA levels were significantly elevated in precursor cell lymphomas and in the poor prognosis group. Elevated ADA was strongly correlated with a poor prognosis. Multivariable analysis identified precursor cell lymphoma, fever, bone marrow involvement, elevated LDH, and elevated ADA as independent predictors of poor prognosis (all p < 0.05). Conclusions: Serum ADA is significantly elevated in pediatric NHL, particularly in precursor cell subtypes and poor prognosis cases, and serves as a potential prognostic marker. ADA may help improve risk stratification and guide personalized treatment strategies. Full article

Propose an enhanced LoRA (Low-Rank Adaptation) MoE (mixed expert) architecture, SLoRA (Enhanced LoRA MoE Architecture), aimed at addressing the key problem of efficient parameter fine-tuning in multitasking scenarios. Given the high cost of traditional full fine-tuning as the parameter size of visual language models increases, and the limitations of LoRA as a popular PEFT (parameter-efficient fine-tuning) method in multitasking, such as inadequate adaptability and difficulty in capturing complex task patterns, as well as the catastrophic forgetting and knowledge fragmentation challenges faced by existing research on integrating mixed expert (MoE) mechanisms into LoRA, SLoRA utilizes orthogonal constraint optimization to reduce disturbance to existing knowledge through constraint solution space initialization, alleviating catastrophic forgetting (old task accuracy retention rate reaches 92.4%, 16.1% higher than LoRA), and an optimized MoE structure that includes general experts (retaining pre-trained knowledge) and task-specific experts (dynamic routing adaptation tasks) to enhance multitask adaptability. Experimental results show that in commonsense reasoning tasks, SLoRA’s accuracy is 9.0% higher than LoRA and 3.7% higher than AdaLoRA on the WSC dataset, and its F1 score is 7.7% higher than LoRA and 2.9% higher than AdaLoRA on the CommonsenseQA dataset; in multimodal tasks, its average score is up to 15.3% higher than LoRA, demonstrating significant advantages over existing methods. Full article

This study developed a transdermal drug delivery system for Rheumatoid Arthritis (RA) using a dual-network hydrogel microneedle patch loaded with methotrexate nanocrystals (DHMN@MTX-NCs), and explored its synergistic therapy with Adalimumab (ADA) for a painless, long-acting, and targeted RA treatment. This study synthesized Methacrylated Hyaluronic Acid and Methacrylated Gelatin. MTX-NCs were prepared by solvent-antisolvent precipitation and incorporated into a dual-network hydrogel microneedle patch via centrifugal molding. Evaluations included pharmaceutical properties, mechanical strength, drug release, in vitro anti-inflammatory effects on RAW 264.7 cells, and therapeutic efficacy in a rat RA model. The experimental results show that the prepared MTX-NCs present a spherical shape, an average size of 325.72 nm, a PDI of 0.154, and a drug-loading capacity of 61.3%. The microneedle patch exhibited high puncture efficiency and suitable swelling. In vitro, DHMN@MTX-NCs combined with ADA most strongly inhibited macrophage migration, upregulated IL-10, and downregulated TNF-α, IL-1β, NO, iNOS, and COX-2. In vivo, both monotherapy and combination therapy reduced joint swelling, bone erosion, and histopathological damage. Ultimately, the study demonstrated the synergistic anti-inflammatory efficacy of DHMN@MTX-NCs combined with ADA, providing a novel, non-invasive, and targeted therapeutic strategy for RA. Full article

With the increasing deployment of autonomous and semi-autonomous road vehicles, Advanced Driver Assistance Systems (ADASs) rely heavily on multi-modal sensing technologies to ensure safe and reliable operation. Among these sensors, Light Detection and Ranging (LiDAR) provides high-resolution three-dimensional environmental perception but is particularly vulnerable to adverse weather conditions such as snowfall. Snowfall can degrade LiDAR performance through signal attenuation, backscattering, false detections, and sensor surface contamination, ultimately reducing visibility and detection reliability. In this study, an experimental investigation was conducted in a climatic chamber to systematically assess LiDAR performance degradation under controlled snowfall conditions. Key parameters influencing sensor behavior, including chamber air temperature, precipitation intensity, and sensor orientation, were isolated and examined. Chamber temperature was varied to generate snow characteristics representative of dry and wet snow, while precipitation intensity was controlled by adjusting snow gun flow rates. Sensor orientation was modified to evaluate its effect on perceived precipitation and snow accumulation. The experimental results confirm the initial hypothesis that snowfall intensity, snow physical properties, and sensor orientation exert a significant influence on LiDAR performance degradation. Increasing precipitation intensity significantly accelerates both 3D target detection loss and 2D visibility reduction, with polynomial regression revealing a non-linear degradation response. Inclined sensor orientations exhibited more rapid performance deterioration compared to a horizontal configuration. These findings provide valuable insights into LiDAR vulnerability in snowy environments and support the development of mitigation strategies to improve ADAS and autonomous vehicle operation in cold climates. Full article

This study examines the confined compressive strength (Fcc) of circular, square, and rectangular column geometries under varying confinement conditions. Results indicate that circular columns have the highest Fcc values, exceeding those of square and rectangular shapes. Increased confinement through clamps significantly enhances compressive strength. Five machine learning models, Linear Regression, Decision Tree, Random Forest, AdaBoost, and Gradient Boosting, were used to predict Fcc based on geometric and confinement parameters. Linear Regression and Decision Tree models achieved moderate predictive performance, with R² values of 0.84 and 0.83, respectively, and relatively higher error measures (RMSE, MAE, and MAPE), indicating limited ability to capture complex nonlinear relationships in the data. In contrast, ensemble-based methods demonstrated superior performance. The Random Forest model improved the coefficient of determination to 0.90 while substantially reducing all error metrics, reflecting enhanced generalization through bagging. The boosting-based approaches yielded the best results, with AdaBoost achieving the highest R² value of 0.99 and the lowest RMSE, MAE, and MAPE among all models, followed closely by Gradient Boosting with an R² of 0.98. These results confirm that ensemble learning techniques, particularly boosting algorithms, yield more accurate and robust predictions than single learners for the problem studied. Data visualization techniques, including Regression Error Characteristic curves (REC) and SHapley Additive exPlanations (SHAP) value analysis, highlighted model performance and feature importance, emphasizing the roles of confinement and geometry in compressive strength. This research demonstrates the potential of machine learning to optimize structural engineering design and suggests further exploration of alternative shapes and confinement strategies to enhance structural integrity. Full article

Murals are vital cultural heritage assets, yet many are increasingly threatened by long-term natural weathering, environmental erosion, and human-induced damage. Given limited conservation resources, an objective method for prioritizing restoration is urgently needed. This study proposes an Individual-Aware Gradient Boosting Regression (IA-GBR) approach to predict the visual saliency of damaged mural regions by integrating physical damage characteristics, spatial location, and observer identity. We construct an eye-tracking dataset containing complete fixation records from multiple participants viewing diverse mural damage types. IA-GBR employs a two-level feature fusion strategy that combines damage, spatial, and individual features within a gradient boosting residual learning framework. The experimental results demonstrate that IA-GBR consistently outperforms baseline methods, including linear and ridge regression, SVR, decision trees, random forests, AdaBoost, and multilayer perceptrons. Feature importance analysis further reveals the relative contributions of individual differences, damage size, spatial position, and semantic factors to saliency formation. The proposed framework provides data-driven support for restoration prioritization and advances perception-aware saliency modeling in cultural heritage conservation. Full article

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disease characterised by the death of motor neurons leading to paralysis and death, generally 3–5 years post-symptom onset. The most frequent genetic cause of ALS is a hexanucleotide repeat expansion (HRE) in the chromosome 9 open reading frame 72 (C9orf72) gene, that has three major hypothesised pathological mechanisms including the production of dipeptide repeat proteins (DPRs). Our laboratory has previously identified purine metabolism dysfunction in induced neural progenitor cell-derived astrocytes (iAstrocytes) from C9orf72 ALS (C9-ALS) cases (C9-iAstrocytes), driven by loss of the enzyme adenosine deaminase (ADA). Here, we have demonstrated that loss of ADA along with changes to ecto-5′-nucleotidase and hypoxanthine-guanine phosphoribosyl transferase led to disruption in purine metabolite levels including purine dNTP output. These changes were recapitulated in patient CSF, whilst loss of ADA was recapitulated in patient white matter. Immunofluorescence also demonstrated purinosome formation dysfunction in C9-iAstrocytes. These changes are likely driven by DPRs as ADA loss was recapitulated in in vitro and in vivo DPR models. Finally, ADA levels could be recovered by reducing DPR levels either by inhibiting serine/arginine-rich splicing factor 1 or overexpressing RuvB-like 2. Our data demonstrate that DPR production negatively affects purine function in C9-ALS suggesting a potentially pivotal role for purine metabolism dysfunction in C9-ALS pathology. Full article

Dry matter intake (DMI) in dairy cows is a critical indicator of nutrient intake from feed, serving as the cornerstone of precision feeding practices, playing a critical role in improving production efficiency and enhancing the quality of dairy products. To address the high costs of traditional measurement methods and the structural complexity and large parameter counts of neural network models, this study proposes a Stacking ensemble learning model to assess DMI, with model parameters optimized using the Tuna Swarm Optimization (TSO) algorithm to enhance assessment accuracy, taking cow body weight, lying duration, lying times, rumination duration, foraging duration, walking steps, and the concentrate-to-roughage feed ratio as input variables. To further improve TSO’s search efficiency and spatial exploration, this study introduces Sine–Logistic chaotic mapping, Levy flight, and Gaussian random walk strategy to optimize the TSO algorithm, developing the improved Tuna Swarm Optimization (ITSO). ITSO-optimized Stacking model achieved superior performance in DMI assessment, with an accuracy of 95.84%, significantly outperforming SVR, RF, DT, GBR, ETR, and AdaBoost models. This study provides a robust tool for precision feeding, contributing to optimizing cow feeding strategies, improving farm efficiency, and supporting sustainable dairy farming practices. Full article

Content/Background: Breast cancer is one of the most fatal cancers among women around the globe. The chances of surviving this cancer increase with early tumor detection, which is necessary for effective treatment. Traditional diagnostic techniques are ineffective and time-consuming, and they yield results that may be accurate or inaccurate. Therefore, our primary objective is to determine how a machine learning model can reduce diagnostic errors and provide accurate results. Objective: The main objective of this project is to build an ML-based classification model that can help doctors detect breast cancer early and more accurately. This project also aims to provide an interactive interface for easy access in healthcare settings. Materials/Methods: For this study, twelve machine learning classification algorithms are implemented and tested: Logistic Regression, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Decision Tree, Random Forest, Gradient Boosting, XGBOOST, Naive Bayes, AdaBoosting, Light GBM, CatBoost, and the Artificial Neural Network (ANN). This study used the Wisconsin Breast Cancer Dataset (WBCD) from the UCI ML Repository. It contains 569 patient samples and 30 features. This dataset has the following features: Radius, Texture, Area, Perimeter, Smoothness, Compactness, Concavity, and Fractional Dimension. The target variable is diagnosis, which is categorized as malignant vs. benign. Results: The fifteen models were analyzed, evaluated, and compared using five performance metrics: Accuracy, Precision, Recall, F1-Score, and AUC-ROC. Among the evaluated models, CatBoost, LoGR, and AdaBoost outperformed the others, with an Accuracy of 97.%, Precision of 97%, Recall of 97%, and AUC-ROC score of 99%. The AUC-ROC is nearly 99%, and the model has a high ability to differentiate between malignant and benign tumors. Full article

Traditional statistical approaches identify group-level associations between biomarkers and rupture status in intracranial aneurysms (IAs) but often miss nonlinear interactions at the patient level. Methods: The authors retrospectively analyzed 35 saccular IAs in 35 patients (57.1% ruptured) from a single center (2021–2023). Demographics, detailed morphology (e.g., neck width, aspect ratio, VERTI, irregular shape), and multi-site inflammatory/immune markers (CRP; complement C3/C4; IgA/IgG/IgM) were included. After preprocessing (min–max scaling; one-hot encoding), five algorithms (DT, AdaBoost, GBM, XGBoost, RF) were evaluated with stratified five-fold CV and class balancing via random oversampling. The primary model (Random Forest) was tuned with Optuna and explained using global feature importance and LIME. The results showed that baseline RF achieved CV ROC-AUC 0.81 and test ROC-AUC 0.92 (test accuracy 0.857). The tuned RF (with oversampling and Optuna) yielded a mean CV accuracy of 0.85 ± 0.09 and CV ROC-AUC of 0.98 ± 0.07 while maintaining test ROC-AUC of 0.92. The average precision on the test PR curve was 0.97. The most influential predictors combined inflammatory markers (CRP, C3, C4) with morphology (neck width, irregular shape). LIME revealed consistent local patterns: low A.CRP/C.CRP and lower C3/C4 favored Not-Broken, whereas higher CRP/complement with smaller neck and irregular shape pushed toward Broken classifications. It can be concluded that an interpretable machine learning (ML) pipeline captured clinically plausible, nonlinear interactions between inflammation and aneurysm geometry. Integrating explainable ML with conventional statistics may enhance rupture risk stratification, enable patient-level rationale, and inform personalized management. These results could significantly contribute to the quality of treatment for patients with intracranial aneurysms. Full article

Current non-intrusive load monitoring (NILM) methods rely on large amounts of labeled historical data and face domain shift issues, which limits the application of deep learning models in practical scenarios. To this end, this paper proposes a SimCLR-ADA-LM framework based on visualized color V-I trajectories. Initially, unlabeled load data from the source domain (PLAID) and target domain (WHITED) are converted into RGB color V-I trajectories and input into the model. The framework enhances intra-class aggregation through contrastive learning and achieves inter-domain feature alignment via adversarial training between the encoder and the domain discriminator to obtain domain-invariant features. Subsequently, the model is fine-tuned using a small amount of labeled data from the target domain to achieve load identification. Ablation and comparative experimental results demonstrate that the proposed model exhibits superior performance advantages over conventional models in cross-domain identification tasks. Furthermore, it maintains significant learning efficiency and recognition robustness even under conditions of limited labeled data. Full article

LiDAR sensors are becoming more common and are going to be widely adopted in vehicles in the future by reducing the production cost of the time-of-flight units. Manufacturers are uncertain about the placement, cover material, and shape of the assembly to achieve the optimal performance of the LiDAR, especially in rainy conditions. Although there are existing methodologies for evaluating the visibility and signal intensity of point clouds, there are no indexing approaches available since they would require a broad and comprehensive dataset and realistic and repeatable conditions to perform parametric studies. A matrix of rain conditions with quantified raindrop distribution characteristics is simulated using a wind tunnel via the wind-driven rain concept to produce the realistic impact of raindrops onto the sensor assembly surface at various wind speeds. This paper presents a performance prediction model method for LiDAR sensors and showcases the capability of such a model to provide insights quantitatively when comparing variations. The model is 3-dimensional, including rain conditions perceived by a moving vehicle at different speeds, material properties of surface wettability, and LiDAR visibility in rain compared to dry conditions. The observed LiDAR signal degradation follows an exponential manner, for which this study provides experimentally derived coefficients, enabling quantitative prediction across materials, topologies, rain, and driving speed conditions. Full article

In complex natural scenarios, facial expression recognition often encounters partial occlusions caused by glasses, hand gestures, and hairstyles, making it difficult for models to extract effective features and thereby reducing recognition accuracy. Existing methods often employ attention mechanisms to enhance expression-related features, but they fail to adequately address the issue where high-frequency responses in occluded regions can disperse attention weights (e.g., incorrectly focus on occluded areas), making it challenging to effectively utilize local cues around the occlusions and limiting performance improvement. To address this, this paper proposes a network based on an adaptive attention mechanism (Adaptive Attention Vision Transformer, AA-ViT). First, an Adaptive Attention module (ADA) is designed to dynamically adjust attention scores in occluded regions, enhancing the effective information in features. Next, a Dual-Branch Multi-Layer Perceptron (DB-MLP) replaces the single linear layer to improve feature representation and model classification capability. Additionally, a Random Erasure (RE) strategy is introduced to enhance model robustness. Finally, to address the issue of model training instability caused by class imbalance in the training dataset, a hybrid loss function combining Focal Loss and Cross-Entropy Loss is adopted to ensure training stability. Experimental results show that AA-ViT achieves expression recognition accuracies of 90.66% and 90.01% on the RAF-DB and FERPlus datasets, respectively, representing improvements of 4.58 and 18.9 percentage points over the baseline ViT model, with only a 24.3% increase in parameter count. Compared to existing methods, the proposed approach demonstrates superior performance in occluded facial expression recognition tasks. Full article