Search Results (3,621)

Background/Objectives: This study aims to develop a deep learning (DL) model integrating ultrasound images and multidimensional clinical information to improve the diagnostic accuracy of breast non-mass lesions (NMLs). Methods: A total of 794 multicenter retrospective cases of NMLs were selected, stratified, and randomly divided into a training set (635 cases) and validation set (159 cases) at an 8:2 ratio. Multidimensional clinical information (including age, reproductive history, menstrual history, medical history, and findings from palpating the lesions) was incorporated to develop a DL model integrating ultrasound images and clinical data. To evaluate the diagnostic performance of the DL model, the area under the curve (AUC), accuracy, specificity, and sensitivity were employed. Results: The diagnostic model for NMLs integrating ultrasound images and multidimensional clinical information achieved an AUC of 0.8520 (95% CI: 0.7898–0.9068), F1 score of 0.7563, accuracy of 0.8176, sensitivity of 0.7031, and specificity of 0.8947. Its performance was superior to that of the model using only ultrasound images (AUC 0.8520 vs. 0.7571). SHAP analysis evaluating the reasons for the improved performance revealed that palpation with indistinct margins, abnormal axillary nodes, and older age were the three features with the highest contribution to predicting malignant risk. Conclusions: The DL model integrating ultrasound images and multidimensional clinical information demonstrated promising diagnostic performance in differentiating benign and malignant breast NMLs, suggesting the complementary value of multidimensional clinical information in the differential diagnosis of NMLs, though the reported AUC of 0.8520 is a preliminary internal estimate that awaits external validation. Full article

The safety of workers in industrial settings is ensured through the correct use of Personal Protective Equipment (PPE). The use of such equipment can be monitored using Deep Learning (DL). Federated Machine Learning (FML) is a technique that can be used in this context to preserve the privacy of sensitive information and provide explainability for the models adopted. Explainability techniques are an essential resource for interpreting the classification performed by the model. In this regard, this study aims to evaluate, through the adoption of specific similarity indices, the robustness and consistency of the explainability algorithms adopted to identify the areas of the images that are decisive for PPE classification. The dataset consists of 1600 real images representing work environments, in which staff are portrayed both with and without Personal Protective Equipment; specifically, there are workers wearing helmets, workers wearing reflective vests, workers wearing both devices and, finally, workers without any PPE. SSIM, VIF and SCC are the most relevant indices involved in the study. In the experimental phase, their mean values stand at 0.99, 0.96 and 0.96 for the intra-client study, and 0.96, 0.91 and 0.71 in the inter-client analysis. Full article

Artificial intelligence (AI) is increasingly used in cancer research, enabling integrative analysis of complex biomedical data to identify actionable therapeutic vulnerabilities. This review specifically examines how AI advances mechanistic cancer target discovery and translational drug development, focusing on: (1) the processing of large-scale genomics, transcriptomics, proteomics, metabolomics, single-cell profiling, spatial, and clinical datasets using machine learning (ML) and deep learning (DL) algorithms; (2) the identification of candidate biomarkers, driver genes, dysregulated pathways, tumor dependencies, and molecular targets that traditional methods often miss; (3) the integration of multi-omics data, network biology, causal inference, and systems-level modeling to refine mechanistic understanding of cancer progression and separate functional driver events from passengers; and (4) applications in drug development, including virtual screening, molecular modeling, structure-informed target validation, drug repurposing, synthetic lethality prediction, and de novo drug design, which collectively may enhance early-stage drug discovery efficiency. The review underscores that AI serves as both a predictive tool and a platform for linking molecular mechanisms to hypothesis generation, target prioritization, and rational treatment design. Challenges such as data heterogeneity, algorithmic bias, interpretability, reproducibility, regulatory requirements, and patient privacy must be addressed for robust translation and clinical use. Future directions may focus on hybrid approaches that integrate causal modeling, explainable AI, multimodal data, and experimental validation to yield mechanistically grounded, clinically actionable insights. AI-driven approaches ultimately aim to accelerate mechanism-based cancer target discovery and enable more precise, biologically informed anticancer therapies. Full article

Multiple sclerosis (MS) is an inflammatory disease of the central nervous system (CNS) that impacts nearly 3 million people worldwide. While the etiology and pathogenesis of MS are not yet fully understood, current evidence suggests that it results from complex interactions between genetic and environmental conditions. Clarifying the autoimmune mechanisms underlying MS remains a central objective in the development of effective therapeutic strategies. Interferon-beta (IFN-$\beta$) is one of the most frequently prescribed disease-modifying treatments for individuals with MS. However, despite its established efficacy, recent studies report that approximately 30–50% of patients exhibit inadequate response to IFN-$\beta$, largely due to genetic variability. Machine learning (ML), a branch of artificial intelligence (AI), employs data-driven computational models to enhance predictive accuracy and classification. In recent MS research, unsupervised learning techniques such as hierarchical clustering and K-means have been applied for classification purposes. However, these methods often fail to yield optimal solutions because they require numerous arbitrary decisions and perform adequately only when datasets contain clusters of similar sizes and lack significant outliers. Fuzzy systems (FSs) are designed to model complex, ambiguous real-world phenomena. In this study, an AI algorithm incorporating a fuzzy system, informed by expert neurologist input, is proposed to enhance the assignment of unknown class labels related to IFN-$\beta$ response in MS patients. Additionally, a genetic algorithm (GA) is introduced to identify optimal solutions within the search space, facilitating hyperparameter optimization of a deep learning (DL) model trained with genetic biomarkers to identify patients likely to benefit from this therapy. Experimental results demonstrate that the fuzzy system achieved $80\%$ classification efficiency, in contrast to $64\%$ with conventional hierarchical clustering. Furthermore, an artificial neural network (ANN) model, with hyperparameters optimized by the GA, achieved an accuracy of 0.8–1.0, surpassing the multi-layer perceptron (MLP), which achieved 0.6–0.8 accuracy using conventional tuning methods. Full article

Power flow (PF) analysis is fundamental for power system operation and planning, yet traditional methods like Newton–Raphson face problems in convergence and computational efficiency. While deep learning (DL) offers promising solutions, its “black-box” nature and unstable training dynamics hinder practical adoption. This paper proposes a physics-informed neural network (PINN) framework integrated with a novel half-inverse gradient (HIG) mechanism to address these limitations. First, a systematic study of gradient scaling in PF optimisation found that the lack of enough inverse matrix compensation was the main cause of training instability. Second, we design a residual-driven HIG method that compensates gradient matrices via inverse operations, enabling accelerated convergence while maintaining numerical stability. Third, we develop parameterized voltage variables with differentiable activation functions to enforce hard operational constraints. The HIG optimizer leverages automatic differentiation and truncated singular value decomposition to balance diagonal/non-diagonal gradient information, achieving 99% accuracy in case4gs and case30 studies. Experiments on case118 demonstrate the framework’s scalability, with 65% accuracy compared to about 38% for baseline physics-informed approaches. Full article

Objectives: To assess the association of structural biomarkers derived from optical coherence tomography (OCT) and choriocapillaris (CC) flow information with point-wise retinal sensitivity (PWS) measured by microperimetry (MP) in intermediate age-related macular degeneration (iAMD). Methods: Patients with iAMD received imaging with spectral-domain (SD)-OCT (Spectralis, Heidelberg Engineering) and OCT-angiography (OCT-A) (PLEX Elite 9000, ZEISS). In addition, MP examinations in photopic setting (MP-3, NIDEK) and mesopic background illumination (MAIA2, ICare) were performed. The thickness of the ellipsoid-zone (EZ) and the outer nuclear layer (ONL), as well as the volume of drusen and HRF, were segmented using deep-learning (DL)-based approaches. CC flow deficit percentage (FD%) was extracted from OCT-A slabs using a novel binarization method. Semiautomatic co-registration of MP examinations, OCT-A slabs, and OCT volumes was performed. Three exploratory models were calculated using multivariable mixed-effects models: (1) structure–function (SF) using structural OCT biomarkers, (2) flow–function (FF) utilizing OCT-A derived flow information, and (3) structure–flow–function (SFF) incorporating both OCT and OCT-A data. Model performance was evaluated using AIC and BIC criterion. Results: 19 eyes of 19 patients were evaluated, totalling 3297 MP-stimuli, 1873 B-scans, and 19 OCT-A slabs. Mean (SD) age was 76 (7) years, and sensitivity was 26.0 (3.36) dB in the MP-3 and 22.42 (3.64) dB in the MAIA2. Mesopic MAIA2 examinations showed significantly lower PWS values (−3.56 to −3.63 dB; p < 0.001). Drusen and HRF volume decreased PWS (−0.6 [95% CI: −1.04; −0.16] dB/nL; p = 0.007 and −9.56 [95% CI: −12.86; −6.26] dB/nL; p < 0.001), while ONL was positively associated with PWS (0.06 [0.05; 0.07] at an eccentricity of 5.2°; p < 0.001) in the SF model. CC FD% was not significantly associated with PWS in the FF and the SFF model (p > 0.05 in both cases). In the SFF model drusen volume (−1.69 [95% CI: −2.09; −1.29] dB/nL; p < 0.001), EZ (0.04 [95% CI: 0.02; 0.06] dB/µm; p < 0.001), and ONL thickness (0.03 [95% CI: 0.02; 0.04] dB/µm; p < 0.001) were significant predictors for PWS. The SF model exhibited the lowest AIC and BIC indicating best model performance. Conclusions: Structural parameters derived from SD-OCT such as HRF, drusen volume, and outer retinal layer thickness may be more closely associated with PWS, with CC FD% as an OCT-A-derived metric contributing limited additional explanatory benefit in cross-sectional analyses. Full article

Colorectal cancer (CRC) remains one of the leading causes of cancer-related mortality worldwide, with prognosis critically dependent on the stage at diagnosis. Traditional tissue biopsy presents well-known limitations, including tumor heterogeneity and invasiveness. Liquid biopsy, encompassing the analysis of circulating tumor DNA (ctDNA), circulating tumor cells (CTCs), exosomes, and other cell-free biomarkers, has emerged as a transformative approach for non-invasive tumor profiling. This comprehensive narrative review outlines the recent evidence published on the current state and future perspectives of liquid biopsy in CRC, with a focused emphasis on the role of artificial intelligence (AI), machine learning (ML), and deep learning (DL) in data analysis and clinical translation. Methods: A narrative review of the literature was conducted by searching PubMed/MEDLINE, EMBASE, and ClinicalTrials.gov for articles published between January 2020 and January 2026, using a predefined Boolean search string combining terms related to liquid biopsy biomarkers, colorectal cancer, and artificial intelligence methodologies. Filters were applied to include only English-language human studies. Additional relevant sources were consulted to ensure comprehensive coverage of the available literature. Liquid biopsy platforms, particularly ctDNA sequencing and methylation profiling, demonstrate increasing clinical utility across the CRC care continuum from population screening to post-surgical minimal residual disease (MRD) detection and real-time therapy monitoring. AI-driven analytical frameworks, including Random Forest, Convolutional Neural Networks, LSTM models, and more recently Large Language Models (LLMs), substantially augment the sensitivity and specificity of liquid biopsy interpretation, enabling multimodal data integration. The convergence of liquid biopsy technology and AI-driven analytics represents a paradigm shift toward precision oncology in CRC. Remaining challenges include analytical standardization, model explainability, regulatory harmonization, and equitable access. Future integration of federated learning frameworks and LLM-based clinical decision support tools will be essential for responsible clinical translation. Full article

Accurate prediction of the El Niño–Southern Oscillation (ENSO) is crucial for understanding and anticipating global climate variability. Although deep learning (DL)-based models have recently improved ENSO forecasting skill, achieving strong predictive performance while maintaining model interpretability remains a major challenge. Existing approaches may suffer from instability or have difficulty distinguishing contributions across multiple variables and time steps. To address this issue, this study presents an interpretable ENSO forecasting framework that combines a ConvNeXt-based deep learning model, ENSO-ConvNeXt, with an improved gradient-based contribution analysis method whose calculation strategy is adjusted according to different ENSO phases. The simplified ConvNeXt architecture facilitates the integration of interpretability methods while retaining strong predictive capability. ENSO-ConvNeXt achieves competitive forecasting skill with an effective lead time exceeding 20 months, accurately capturing the Niño3.4 index evolution during the peak season and the temporal evolution of ENSO events. The case studies of representative ENSO events demonstrate that the major contribution regions identified by the model are broadly consistent with established ENSO variability patterns across major ocean basins. These results highlight the potential of our framework to advance ENSO prediction while providing statistically grounded and physically interpretable insights. Full article

Recent advancements in 3D reconstruction technologies have significantly transformed plant phenotyping, enabling precise, scalable, and automated trait extraction. Traditional manual phenotyping methods are increasingly being replaced by image-based approaches, such as photogrammetry, LiDAR, RGB-D sensing, and deep learning (DL)-based techniques. These tools allow for non-destructive, high-throughput measurements of plant morphology, structure, and physiological traits. This review synthesizes the state of the art in 3D reconstruction methods, including conventional geometric algorithms and emerging DL methods, and evaluates their application across diverse plant species. In addition, we discuss the sensing modalities, evaluation metrics, and crop-specific deployments. Although promising, current technologies still face challenges in terms of computational efficiency, scalability to outdoor environments, and generalizability across crop types. This review concludes by identifying research gaps and future directions for making real-time, field-deployable 3D phenotyping systems. Full article

Interferometric Synthetic Aperture Radar (InSAR) enables regional monitoring of ground deformation, but operational geohazard analysis remains challenged by atmospheric artefacts, temporal decorrelation, and the need for scalable interpretation of multi-temporal products. A systematic review was conducted through searches in Scopus and Web of Science, resulting in 135 peer-reviewed scientific articles on the integration of Machine Learning (ML) and Deep Learning (DL) with multi-temporal InSAR (MT-InSAR). The literature is dominated by applications to landslides and land subsidence, with additional studies addressing volcanic unrest and other deformation-related hazards. Persistent Scatterer (PS) and Small-Baseline Subset (SBAS) approaches are frequently used to derive deformation time series, which are then coupled with ML/DL for the detection and mapping of active phenomena and for short-horizon forecasting. Convolutional architectures, such as Convolutional Neural Networks (CNNs), are commonly reported for spatial recognition tasks, while recurrent models like Long Short-Term Memory (LSTM) networks are often applied to time-series prediction. Reported benefits include improved automation and predictive performance, although sensitivity to noise sources remains a challenge. Overall, the evidence supports AI-enabled InSAR workflows for scalable geohazard monitoring, while highlighting the need for standardized benchmarks and systematic transferability assessment. This review provides a roadmap for transitioning from research prototypes to operational early-warning systems. Full article

Background/Objectives: Prostate cancer (PCa) is the second most commonly diagnosed cancer in men globally. Radiation oncologists often find PCa tumor characterization and outcome prediction challenging. Therefore, the potential for artificial intelligence (AI) implementation in radiation oncology has increased in recent years. This systematic review aims to evaluate the efficacy of AI algorithms in characterizing PCa tumors and predicting post-therapy outcomes. Methods: A total of 2055 studies were identified through a comprehensive search across PubMed and Scopus, then exported to Covidence. Inclusion criteria focused on prospective and retrospective cohort studies as well as randomized clinical trials (RCTs) published between 2015 and 2024 that explored the implementation of AI in tumor characterization and outcome prediction of PCa. Two independent reviewers evaluated each paper, and evaluation metrics such as specificity, sensitivity, accuracy, and area under the curve (AUC) were analyzed. The Risk of Bias in Non-randomized Studies of Interventions, Version 2 (ROBINS-I V2) tool was used to assess the risk of bias (ROB). Results: Across the 19 studies analyzed, there was no significant difference in model performance between machine learning (ML) and deep learning (DL) models. AI models using multi-input strategies (e.g., radiomics with clinical markers) generally performed better than single-input models. Of the imaging modalities used for radiomic feature extraction, multiparametric MRI (mpMRI)-trained AI models consistently achieved the highest performance. Conclusions: AI displays considerable potential for integration into clinical workflows for PCa management. However, further studies utilizing larger datasets and external cohorts independent of the sample population are needed to validate clinical utility and improve model transparency for reliable implementation. Full article

Background/Objectives: Brain tumors are highly lethal cancers, with gliomas representing the most complex subtype. Magnetic resonance imaging (MRI) is the main non-invasive imaging modality. This review evaluates deep learning (DL) and artificial intelligence methods for brain tumor segmentation and classification. Methods: In this systematic review, PubMed and Scopus were searched for articles published from 2022 to March 2025. Authors independently identified eligible studies based on predefined inclusion criteria and extracted data. The study quality and risk of bias were assessed using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) checklist. Results: Thirty-one studies met the inclusion criteria from 310 records, with eight addressing both segmentation and classification. Most segmentation studies used publicly available multiparametric MRI datasets. Performance varied by architecture and tumor region, with whole-tumor segmentation achieving the highest Dice Similarity Coefficient (DSC). Classical U-Nets reported DSC values ranging 80–87%, while models with residual or attention mechanisms exceeded 90%. Classification focused on tumor type and glioma grading, using features learned from multiparametric MRI. Reported accuracy ranged from 91.3% to 99.4%, with sensitivity and specificity often above 95%. However, variability across tumor subregions, limited external validation, reliance on public datasets, and heterogeneous preprocessing raise concerns about robustness and real-world generalizability. Evidence on the use of explainability methods for both tasks remains limited. Conclusions: DL models for glioma segmentation and classification demonstrate promising performance. However, standardized validation protocols, multi-center datasets, and the integration of explainable artificial intelligence techniques are needed to improve transparency, robustness, and clinical applicability. Full article

Battery state-of-health (SOH) estimation is essential for the safety and reliability of energy storage systems. However, incomplete measurements due to sensor or communication failures pose significant challenges for accurate prediction. This paper proposes a robust SOH estimation framework using a minimal 5 min observation window to handle high data sparsity in both random and latter-half missing scenarios. Three Deep Learning (DL) architectures—Long Short-Term Memory (LSTM), Bidirectional LSTM (BiLSTM), and Transformer—are evaluated for data imputation and SOH estimation against traditional polynomial fitting. Simulation results on the NASA benchmark dataset demonstrate that the proposed LSTM model achieves high accuracy, with an RMSE of 0.8522 on complete data. For imperfect data scenarios, BiLSTM-based imputation effectively suppresses extreme deviations, reducing the Maximum Error (MxE) by 44% (from 14.04 to 7.85) compared to traditional polynomial methods. Furthermore, in challenging terminal missing-data cases, a hybrid LSTM-Transformer strategy maintains physical consistency, achieving a superior RMSE of 1.0026. These findings confirm that the proposed DL-based framework significantly outperforms conventional techniques, providing a robust and reliable solution for real-time battery health monitoring under unpredictable data conditions. Full article

E-commerce web content is dynamic and diverse, necessitating continuous monitoring and adaptation. This presents researchers with the challenge of discovering methods to improve delivered services. Hence, integrating natural language processing (NLP), Machine Learning (ML), Deep Learning (DL), and sentiment analysis (SA) provides businesses with robust frameworks to utilize customer feedback and enhance decision-making. Therefore, we introduce a novel dataset collection methodology that captures the dynamic relationships between e-commerce web content and consumer sentiment. Additionally, we introduce a novel, real-consumer-based quality metric on product content through FG-CSrP, extending SA into a new Fine-Grained Consumer Sentiment related to the Product. We evaluated our dataset using baseline models: Deep Neural Network (DNN), Long Short-Term Memory (LSTM), DistilBERT, and twelve automatically optimized models created by AutoGluon-Tabular across three scenarios, each with varying feature inputs (numerical, textual, and both). We then applied Explainable Artificial Intelligence (XAI) to the DNN model to explain feature importance in prediction. Our findings showed that LightGBMXT outperformed the other models in two out of three scenarios, and XAI interpretations highlighted the significant role of vendor-provided web content details in consumer sentiment. Overall, our approach provides actionable insights that can help vendors improve e-commerce strategies and strengthen customer engagement. Full article

Brain tumors (BTs) arise from the abnormal growth of cells within brain tissue and may spread rapidly, making them a major cause of mortality worldwide. Early detection of BTs remains highly challenging due to the brain’s complex structure and the heterogeneous nature of tumors. Magnetic Resonance Imaging (MRI) provides detailed information about tumor size, location, and shape, thereby supporting clinical decision-making for treatments such as chemotherapy, radiation therapy, and surgery. Traditional machine learning (ML) approaches mainly rely on manual feature extraction, whereas recent advances in Computer-Aided Diagnosis (CAD) and deep learning (DL) have enabled more accurate detection of small and complex tumor regions. To improve automated tumor detection, we propose a hybrid Swin–YOLO framework that combines the Swin Transformer (ST) with the latest CNN-based YOLOv12 model. In this framework, the Swin Transformer serves as the main backbone for feature extraction, while the Feature Pyramid Network (FPN) and Path Aggregation Network (PANet) are employed in the neck to better capture multi-scale features. For training, we used the publicly available Br35H dataset and applied data augmentation to enhance the model’s robustness and generalization capability. The experimental results show that the proposed framework achieved 99.7% accuracy, 99.4% mAP@50, and 87.2% mAP@50:95. Furthermore, we incorporated Explainable Artificial Intelligence (XAI) techniques, including Grad-CAM and SHAP, to improve the interpretability of the model by visually highlighting the tumor regions that contributed most to the prediction. In addition, we developed NeuroVision AI, a web-based application designed to support faster and more accurate clinical decision-making. Although the proposed model demonstrated strong performance on the dataset, these results should be interpreted within the context of the current experimental setting. Full article