Search Results (1,562)

Accurate melt pool geometry prediction is essential for ensuring quality and reliability in Laser Powder Bed Fusion (L-PBF). However, small experimental datasets and limited physical interpretability often restrict the effectiveness of traditional machine learning (ML) models. This study proposes a hybrid framework that integrates an explicit thermal model with ML algorithms to improve prediction under sparse data conditions. The explicit model—calibrated for variable penetration depth and absorptivity—generates synthetic melt pool data, augmenting 36 experimental samples across conduction, transition, and keyhole regimes for 316 L stainless steel. Three ML methods—Multilayer Perceptron (MLP), Random Forest, and XGBoost—are trained using fivefold cross-validation. The hybrid approach significantly improves prediction accuracy, especially in unstable transition regions (D/W ≈ 0.5–1.2), where morphological fluctuations hinder experimental sampling. The best-performing model (MLP) achieves R² > 0.98, with notable reductions in MAE and RMSE. The results highlight the benefit of incorporating physically consistent, nonlinearly distributed synthetic data to enhance generalization and robustness. This physics-augmented learning strategy not only demonstrates scientific novelty by integrating mechanistic modeling into data-driven learning, but also provides a scalable solution for intelligent process optimization, in situ monitoring, and digital twin development in metal additive manufacturing. Full article

Environmental noise is inevitable during seismic data acquisition, with major sources including heavy machinery, rivers, wind, and other environmental factors. During field data acquisition, it is important to assess the impact of environmental noise and evaluate data quality. In subsequent seismic data processing, these noise components also need to be eliminated. Accurate identification of noise traces facilitates rapid quality control (QC) during fieldwork and provides a reliable basis for targeted noise attenuation. Conventional environmental noise identification primarily relies on amplitude differences. However, in seismic data, high-amplitude signals are not necessarily caused by environmental noise. For example, surface waves or traces near the shot point may also exhibit high amplitudes. Therefore, relying solely on amplitude-based criteria has certain limitations. To improve noise identification accuracy, we use the Mel-spectrogram to extract features from seismic data and construct the dataset. Compared to raw time-series signals, the Mel-spectrogram more clearly reveals energy variations and frequency differences, helping to identify noise traces more accurately. We then employ a Vision Transformer (ViT) network to train a model for identifying noise in seismic data. Tests on synthetic and field data show that the proposed method performs well in identifying noise. Moreover, a denoising case based on synthetic data further confirms its general applicability, making it a promising tool in seismic data QC and processing workflows. Full article

Large, high-quality patient datasets are essential for applications like economic modeling and patient simulation. However, real-world data is often inaccessible or incomplete. Synthetic patient data offers an alternative, and current methods often fail to preserve clinical plausibility, real-world correlations, and logical consistency. This study presents a patient cohort generator designed to produce realistic, statistically valid synthetic datasets. The generator uses predefined probability distributions and Cholesky decomposition to reflect real-world correlations. A dependency matrix handles variable relationships in the right order. Hard limits block unrealistic values, and binary variables are set using percentiles to match expected rates. Validation used two datasets, NHANES (2021–2023) and the Framingham Heart Study, evaluating cohort diversity (general, cardiac, low-dimensional), data sparsity (five correlation scenarios), and model performance (MSE, RMSE, R², SSE, correlation plots). Results demonstrated strong alignment with real-world data in central tendency, dispersion, and correlation structures. Scenario A (empirical correlations) performed best (R² = 86.8–99.6%, lowest SSE and MAE). Scenario B (physician-estimated correlations) also performed well, especially in a low-dimensions population (R² = 80.7%). Scenario E (no correlation) performed worst. Overall, the proposed model provides a scalable, customizable solution for generating synthetic patient cohorts, supporting reliable simulations and research when real-world data is limited. While deep learning approaches have been proposed for this task, they require access to large-scale real datasets and offer limited control over statistical dependencies or clinical logic. Our approach addresses this gap. Full article

Digital image quality is crucial for reliable analysis in applications such as medical imaging, satellite remote sensing, and video surveillance. However, traditional denoising methods struggle to balance noise removal with detail preservation and lack adaptability to various types of noise. We propose a novel three-module architecture for image denoising, comprising a generator, a dual-path-UNet-based denoiser, and a discriminator. The generator creates synthetic noise patterns to augment training data, while the dual-path-UNet denoiser uses multiple receptive field modules to preserve fine details and dense feature fusion to maintain global structural integrity. The discriminator provides adversarial feedback to enhance denoising performance. This dual-path adversarial training mechanism addresses the limitations of traditional methods by simultaneously capturing both local details and global structures. Experiments on the SIDD, DND, and PolyU datasets demonstrate superior performance. We compare our architecture with the latest state-of-the-art GAN variants through comprehensive qualitative and quantitative evaluations. These results confirm the effectiveness of noise removal with minimal loss of critical image details. The proposed architecture enhances image denoising capabilities in complex noise scenarios, providing a robust solution for applications that require high image fidelity. By enhancing adaptability to various types of noise while maintaining structural integrity, this method provides a versatile tool for image processing tasks that require preserving detail. Full article

Flocculants are indispensable in water and wastewater treatment, enabling the aggregation and removal of suspended particles, colloids, and emulsions. However, the conventional development and application of flocculants rely heavily on empirical methods, which are time-consuming, resource-intensive, and environmentally problematic due to issues such as sludge production and chemical residues. Recent advances in machine learning (ML) have opened transformative avenues for the design, optimization, and intelligent application of flocculants. This review systematically examines the integration of ML into flocculant research, covering algorithmic approaches, data-driven structure–property modeling, high-throughput formulation screening, and smart process control. ML models—including random forests, neural networks, and Gaussian processes—have successfully predicted flocculation performance, guided synthesis optimization, and enabled real-time dosing control. Applications extend to both synthetic and bioflocculants, with ML facilitating strain engineering, fermentation yield prediction, and polymer degradability assessments. Furthermore, the convergence of ML with IoT, digital twins, and life cycle assessment tools has accelerated the transition toward sustainable, adaptive, and low-impact treatment technologies. Despite its potential, challenges remain in data standardization, model interpretability, and real-world implementation. This review concludes by outlining strategic pathways for future research, including the development of open datasets, hybrid physics–ML frameworks, and interdisciplinary collaborations. By leveraging ML, the next generation of flocculant systems can be more effective, environmentally benign, and intelligently controlled, contributing to global water sustainability goals. Full article

As voice cloning technology rapidly advances, the risk of personal voices being misused by malicious actors for fraud or other illegal activities has significantly increased, making the collection of speech data increasingly challenging. To address this issue, this study proposes a data augmentation method based on XText-to-Speech (XTTS) synthesis to tackle the challenges of small-sample, multi-class speech recognition, using profanity as a case study to achieve high-accuracy keyword recognition. Two models were therefore evaluated: a CNN model (Proposed-I) and a CNN-Transformer hybrid model (Proposed-II). Proposed-I leverages local feature extraction, improving accuracy on a real human speech (RHS) test set from 55.35% without augmentation to 80.36% with XTTS-enhanced data. Proposed-II integrates CNN’s local feature extraction with Transformer’s long-range dependency modeling, further boosting test set accuracy to 88.90% while reducing the parameter count by approximately 41%, significantly enhancing computational efficiency. Compared to a previously proposed incremental architecture, the Proposed-II model achieves an 8.49% higher accuracy while reducing parameters by about 98.81% and MACs by about 98.97%, demonstrating exceptional resource efficiency. By utilizing XTTS and public corpora to generate a novel keyword speech dataset, this study enhances sample diversity and reduces reliance on large-scale original speech data. Experimental analysis reveals that an optimal synthetic-to-real speech ratio of 1:5 significantly improves the overall system accuracy, effectively addressing data scarcity. Additionally, the Proposed-I and Proposed-II models achieve accuracies of 97.54% and 98.66%, respectively, in distinguishing real from synthetic speech, demonstrating their strong potential for speech security and anti-spoofing applications. Full article

Objectives: Automatic tip localization is critical in ultrasound (US)-guided interventions. Although deep learning (DL) has been widely used for precise tip detection, existing methods are limited by the availability of real puncture data and expert annotations. Methods: To address these challenges, we propose a novel method that uses synthetic US puncture data to pre-train DL-based tip detectors, improving their generalization. Synthetic data are generated by fusing clinical US images of healthy controls with tips created using generative DL models. To ensure clinical diversity, we constructed a dataset from scans of 20 volunteers, covering 20 organs or anatomical regions, obtained with six different US machines and performed by three physicians with varying expertise levels. Tip diversity is introduced by generating a wide range of synthetic tips using a denoising probabilistic diffusion model (DDPM). This method synthesizes a large volume of diverse US puncture data, which are used to pre-train tip detectors, followed by subsequently training with real puncture data. Results: Our method outperforms MSCOCO pre-training on a clinical puncture dataset, achieving a 1.27–7.19% improvement in AP_0.1:0.5 with varying numbers of real samples. State-of-the-art detectors also show performance gains of 1.14–1.76% when applying the proposed method. Conclusions: The experimental results demonstrate that our method enhances the generalization of tip detectors without relying on expert annotations or large amounts of real data, offering significant potential for more accurate visual guidance during US-guided interventions and broader clinical applications. Full article

Internal crack defects in high-strength low-alloy (HSLA) steels during continuous casting pose significant challenges to downstream processing and product reliability. However, due to the inherent class imbalance in industrial defect datasets, conventional machine learning models often suffer from poor sensitivity to minority class instances. This study proposes a predictive framework that integrates conditional tabular generative adversarial network (CTGAN) for synthetic minority sample generation and CatBoost for classification. A dataset of 733 process records was collected from a continuous caster, and 25 informative features were selected using mutual information. CTGAN was employed to augment the minority class (crack) samples, achieving a balanced training set. Feature distribution analysis and principal component visualization indicated that the synthetic data effectively preserved the statistical structure of the original minority class. Compared with the other machine learning methods, including KNN, SVM, and MLP, CatBoost achieved the highest metrics, with an accuracy of 0.9239, precision of 0.9041, recall of 0.9018, and F1-score of 0.9022. Results show that CTGAN-based augmentation improves classification performance across all models. These findings highlight the effectiveness of GAN-based augmentation for imbalanced industrial data and validate the CTGAN–CatBoost model as a robust solution for online defect prediction in steel manufacturing. Full article

In recent years, data augmentation techniques have become the predominant approach for addressing highly imbalanced classification problems in machine learning. Algorithms such as the Synthetic Minority Over-sampling Technique (SMOTE) and Conditional Tabular Generative Adversarial Network (CTGAN) have proven effective in synthesizing minority class samples. However, these methods often introduce distributional bias and noise, potentially leading to model overfitting, reduced predictive performance, increased computational costs, and elevated cybersecurity risks. To overcome these limitations, we propose a novel architecture, FADEL, which integrates feature-type awareness with a supervised discretization strategy. FADEL introduces a unique feature augmentation ensemble framework that preserves the original data distribution by concurrently processing continuous and discretized features. It dynamically routes these feature sets to their most compatible base models, thereby improving minority class recognition without the need for data-level balancing or augmentation techniques. Experimental results demonstrate that FADEL, solely leveraging feature augmentation without any data augmentation, achieves a recall of 90.8% and a G-mean of 94.5% on the internal test set from Kaohsiung Chang Gung Memorial Hospital in Taiwan. On the external validation set from Kaohsiung Medical University Chung-Ho Memorial Hospital, it maintains a recall of 91.9% and a G-mean of 86.7%. These results outperform conventional ensemble methods trained on CTGAN-balanced datasets, confirming the superior stability, computational efficiency, and cross-institutional generalizability of the FADEL architecture. Altogether, FADEL uses feature augmentation to offer a robust and practical solution to extreme class imbalance, outperforming mainstream data augmentation-based approaches. Full article

Medical imaging is vital for diagnosing and treating colorectal cancer (CRC), a leading cause of mortality. Classifying colorectal polyps and CRC precursors remains challenging due to operator variability and expertise dependence. Deep learning (DL) models show promise in polyp classification but face adoption barriers due to their ‘black box’ nature, limiting interpretability. This study presents an example-based explainable artificial intehlligence (XAI) approach using Pix2Pix to generate synthetic polyp images with controlled size variations and LIME to explain classifier predictions visually. EfficientNet and Vision Transformer (ViT) were trained on datasets of real and synthetic images, achieving strong baseline accuracies of 94% and 96%, respectively. Image quality was assessed using PSNR (18.04), SSIM (0.64), and FID (123.32), while classifier robustness was evaluated across polyp sizes. Results show that Pix2Pix effectively controls image attributes like polyp size despite limitations in visual fidelity. LIME integration revealed classifier vulnerabilities, underscoring the value of complementary XAI techniques. This enhances DL model interpretability and deepens understanding of their behaviour. The findings contribute to developing explainable AI tools for polyp classification and CRC diagnosis. Future work will improve synthetic image quality and refine XAI methodologies for broader clinical use. Full article

This study investigates the parameters influencing the compatibility between cement and polycarboxylate ether (PCE) admixtures in cements produced with various types and dosages of grinding aids (GAs). A total of 29 cement types (including a control) were prepared using seven different GAs at four dosage levels, and 87 paste mixtures were produced with three PCE dosages. Rheological behavior was evaluated via the Herschel–Bulkley model, focusing on dynamic yield stress (DYS) and viscosity. The data were modeled using CNN, Random Forest (RF), and Neural Classification and Regression Tree (NCART), and each model was enhanced with synthetic data generated by Large Language Models (LLMs), resulting in CNN-LLM, RF-LLM, and NCART-LLM variants. All six variants were evaluated using R-squared, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Logcosh. This study is among the first to use LLMs for synthetic data augmentation. It augmented the experimental dataset synthetically and analyzed the effects on the study results. Among the baseline methods, NCART achieved the best performance for both viscosity (MAE = 1.04, RMSE = 1.33, R² = 0.84, Logcosh = 0.57) and DYS (MAE = 8.73, RMSE = 11.50, R² = 0.77, Logcosh = 8.09). Among baseline models, NCART performed best, while LLM augmentation significantly improved all models’ predictive accuracy. It was also observed that cements produced with GA exhibited higher DYS and viscosity than the control, likely due to finer particle size distribution. Overall, the study highlights the potential of LLM-based synthetic augmentation in modeling cement admixture compatibility. Full article

The increasing complexity and scale of electronic health records (EHRs) demand advanced tools for efficient data retrieval, summarization, and comparative analysis in clinical practice. MERA (Medical Electronic Records Assistant) is a Retrieval-Augmented Generation (RAG)-based AI system that addresses these needs by integrating domain-specific retrieval with large language models (LLMs) to deliver robust question answering, similarity search, and report summarization functionalities. MERA is designed to overcome key limitations of conventional LLMs in healthcare, such as hallucinations, outdated knowledge, and limited explainability. To ensure both privacy compliance and model robustness, we constructed a large synthetic dataset using state-of-the-art LLMs, including Mistral v0.3, Qwen 2.5, and Llama 3, and further validated MERA on de-identified real-world EHRs from the MIMIC-IV-Note dataset. Comprehensive evaluation demonstrates MERA’s high accuracy in medical question answering (correctness: 0.91; relevance: 0.98; groundedness: 0.89; retrieval relevance: 0.92), strong summarization performance (ROUGE-1 F1-score: 0.70; Jaccard similarity: 0.73), and effective similarity search (METEOR: 0.7–1.0 across diagnoses), with consistent results on real EHRs. The similarity search module empowers clinicians to efficiently identify and compare analogous patient cases, supporting differential diagnosis and personalized treatment planning. By generating concise, contextually relevant, and explainable insights, MERA reduces clinician workload and enhances decision-making. To our knowledge, this is the first system to integrate clinical question answering, summarization, and similarity search within a unified RAG-based framework. Full article

Background/Objectives: X-ray computed tomography (XCT) is a powerful tool for volumetric imaging, where three-dimensional (3D) images are generated from a large number of individual X-ray projection images. However, collecting the required number of low-noise projection images is time-consuming, limiting its applicability to scenarios requiring high temporal resolution, such as the study of dynamic processes. Inspired by stereo vision, we previously developed stereo X-ray imaging methods that operate with only two X-ray projections, enabling the 3D reconstruction of point and line fiducial markers at significantly faster temporal resolutions. Methods: Building on our prior work, this paper demonstrates the use of stereo X-ray techniques for 3D reconstruction of sharp object corners, eliminating the need for internal fiducial markers. This is particularly relevant for deformation measurement of manufactured components under load. Additionally, we explore model training using synthetic data when annotated real data is unavailable. Results: We show that the proposed method can reliably reconstruct sharp corners in 3D using only two X-ray projections. The results confirm the method’s applicability to real-world stereo X-ray images without relying on annotated real training datasets. Conclusions: Our approach enables stereo X-ray 3D reconstruction using synthetic training data that mimics key characteristics of real data, thereby expanding the method’s applicability in scenarios with limited training resources. Full article

Accurate segmentation of cardiac MR images using deep neural networks is crucial for cardiac disease diagnosis and treatment planning, as it provides quantitative insights into heart anatomy and function. However, achieving high segmentation accuracy relies heavily on extensive, precisely annotated datasets, which are costly and time-consuming to obtain. This study addresses this challenge by proposing a novel data augmentation framework based on a condition-guided diffusion generative model, controlled by multiple cardiac labels. The framework aims to expand annotated cardiac MR datasets and significantly improve the performance of downstream cardiac segmentation tasks. The proposed generative data augmentation framework operates in two stages. First, a Label Diffusion Module is trained to unconditionally generate realistic multi-category spatial masks (encompassing regions such as the left ventricle, interventricular septum, and right ventricle) conforming to anatomical prior probabilities derived from noise. Second, cardiac MR images are generated conditioned on these semantic masks, ensuring a precise one-to-one mapping between synthetic labels and images through the integration of a spatially-adaptive normalization (SPADE) module for structural constraint during conditional model training. The effectiveness of this augmentation strategy is demonstrated using the U-Net model for segmentation on the enhanced 2D cardiac image dataset derived from the M&M Challenge. Results indicate that the proposed method effectively increases dataset sample numbers and significantly improves cardiac segmentation accuracy, achieving a 5% to 10% higher Dice Similarity Coefficient (DSC) compared to traditional data augmentation methods. Experiments further reveal a strong correlation between image generation quality and augmentation effectiveness. This framework offers a robust solution for data scarcity in cardiac image analysis, directly benefiting clinical applications. Full article

Background/Objectives: The aim of this study was to evaluate the impact of inhomogeneity correction (IC) of dose distribution on the dosimetric and radiobiological efficacy of radiation treatment for heterotopic ossification (HO). Methods: This study involved a retrospective analysis of 21 patients treated using a homogeneous dose distribution plan for hip prophylactic HO. These IC-off plans were evaluated against an IC-on dose distribution plan. Dosimetric and corresponding radiobiological parameters (gEUD, LQ-EUD, LQ, EQD2 for α/β = 3 and 10 Gy) were calculated. These parameters were compared for both treatment plans. Additionally, Monte Carlo simulations were performed using mean and standard deviation values from baseline data to generate 10,000 synthetic datasets, allowing for robust statistical modeling of variability in dose distributions and biological outcomes. Results: The homogeneous (IC-off) plans demonstrated overestimation of dose conformity and uniformity, reflected in lower HI values (0.10 ± 0.05 vs. 0.18 ± 0.05) and higher D_90%–D_98% coverage. Radiobiologically, these plans yielded higher gEUD (7.02 Gy vs. 6.80 Gy) and EQD2 values across all α/β scenarios (e.g., EQD2_{[α/β=3]_gEUD} = 14.07 Gy vs. 13.35 Gy), with statistically significant differences (p < 0.001). Although IC-on plans demonstrated steeper dose gradients (higher GIs), this came at the expense of internal dose variability and potentially compromised biological effectiveness. Conclusions: Our results suggest that plans without IC deliver suboptimal biological effectiveness if continued preferentially in routine HO prophylaxis. With advanced radiation dose calculation algorithms available in all centers, inhomogeneity-corrected doses warrant prospective validation. Full article