A molecular reappraisal of quilty lesions: Insights from tissue and circulating biomarkers in heart transplantation.

Battery aging and behavioral pattern identification: A fleet analytics framework for regulatory compliance testing

Improving short text clustering through hybrid augmentation and contrastive learning

Abstract Simultaneous Localization and Mapping (SLAM) is essential for autonomous navigation and positioning of mobile robots. However, most existing LiDAR-based SLAM methods rely on specific environmental assumptions and observation models, which significantly degrade robustness under different degenerate scenarios. To address this issue, this research presents a multi-module degeneracy-resistant SLAM framework (MDR-SLAM) designed to enhance robustness in diverse degenerate scenarios. In the front-end, a clustering and segmentation model based on the variation of point cloud density is designed to filter dynamic interference and invalid point clouds. Additionally, a B-spline-based motion error correction method is proposed to correct IMU errors by fitting two continuous time trajectories within a sliding window. In the degeneracy-handling stage, a degeneracy detection and decoupling method based on features is proposed. By evaluating the feature distribution and the constraint ability of the LiDAR system, the degenerate state is identified and further decoupled into rotational and translational degeneracy factors, The motion state of IMU is then used to compensate for the degenerate motion state of the LiDAR in a targeted manner. In the back-end, degeneracy factors are introduced into the iterative extended Kalman filter, working with the observation model to suppress abnormal observation data. Finally, experiments in diverse degenerate scenarios, such as feature scarcity, dynamic interference, structural repetition, and platform instability, show the proposed method achieves the smallest pose errors (APE and RPE) and the highest-quality mapping results in most scenarios involving different types of degeneracy, validating its adaptability to diverse degenerate scenarios.

Non-negative matrix factorization directly decomposes data features into a base matrix and community matrix, which are easily affected by noise. Multi-view datasets have multiple feature matrices, each with a different angle. The data features need to be re-synthesized rather than simply concatenated or added. Based on the advantages and disadvantages of multi-view clustering and non-negative matrix factorization, we attempt to transplant the method of analyzing abstract connected graphs, analogize the similarity between edges and samples in the graph, and propose a deep non-negative matrix factorization model for clustering by constructing a similarity matrix and decomposing it. At the same time, in order to reduce the interference of noise, we introduce a denoising auto-encoder and non-negative matrix factorization in series, and research the reconstruction features, ultimately forming a model structure framework of “denoising auto-encoder, non-negative matrix factorization, clustering”. Through experiments, the denoising auto-encoder-enhanced non-negative matrix factorization achieved good results on five datasets. It achieved an accuracy of 87 percenton the BBC Sport dataset and 61 percent on Wiki-fea, which increased by two percentage points. The clustering results demonstrate that the model can effectively alleviate the impact of noise and provide new ideas for how to integrate multi-view features.

Deep multi-view clustering aims to exploit the rich semantic information contained in heterogeneous multi-view data to uncover the underlying relationships among samples. However, existing deep multi-view clustering models often overlook intercluster separability and the effective integration of semantic information across views, resulting in insufficient feature discriminability and consequently limited clustering performance. To address the above issues, this paper proposes a novel deep multi-view clustering method via cluster-semantic guidance. We separate clusters to enhance inter-cluster discriminability, while incorporating a knowledge distillation mechanism to ensure cluster stability and facilitate the learning of clustering-friendly representations. Furthermore, by aggregating sample-level semantic information, the model is guided to follow a cluster-oriented learning strategy that promotes the extraction of discriminative features, thereby strengthening the sample representation capability. Our method effectively learns discriminative and clustering-friendly representations, guiding the model to acquire distinctive feature embeddings from a cluster-oriented perspective. Our comprehensive experiments across datasets of varying scales confirm the model's effectiveness, showing superior clustering performance over existing state-of-the-art methods.

ABSTRACT Purpose : This study aimed to determine whether routinely collected clinical data from a university dental clinic could be translated into a coherent framework for organizing competency‐based clinical training. By examining patterns of patient complexity, the study sought to generate an evidence‐informed set of educational care lines to guide case allocation, support competency development, and strengthen assessment practices. Methods : A total of 331 anonymized patient records from 2023 to 2025 were extracted from the institutional database. Variables included demographic information, reasons for appointments, treatment duration, number of visits, procedure codes, and completion status. After data cleaning, relevant variables were selected based on their educational relevance. A multivariate K ‐means clustering model ( K = 2–7) was applied using standardized numerical variables and one‐hot‐encoded categorical variables. The optimal solution ( K = 2) was identified through silhouette analysis. Cluster profiles were examined and interpreted pedagogically to generate educational care lines. Results : Two distinct macro‐clusters emerged, reflecting low and high clinical complexity. Low‐complexity patients ( n = 282) typically underwent short, straightforward treatments with high completion rates. High‐complexity patients ( n = 49) demonstrated longer treatment trajectories, multiple procedures, and a higher risk of interruption. These patterns informed the derivation of five educational care lines: preventive care, simple restorative care, complex chronic care, prosthodontic care, and critical adherence care. For each line, corresponding competencies, learning objectives, assessment criteria, and autonomy expectations were defined. Conclusions : This proof‐of‐concept study demonstrates that clinical data can be transformed into a structured educational framework capable of informing competency‐based curriculum design. The resulting care‐line model offers a practical method for aligning case complexity with student readiness, improving consistency in clinical exposure, and supporting more reliable assessment practices. Further validation in larger or multicenter cohorts is warranted.

With the rapid growth of e-commerce and increasing competition in the express logistics industry, effective customer management has become critical for improving operational efficiency and revenue stability. This study proposes a data-driven framework for refined customer management based on real-world waybill data. Customer behavior is characterized using multi-dimensional features derived from aggregated transaction records. A hybrid segmentation approach combining a customer value four-quadrant model and K-means clustering is employed to identify customer value levels and behavioral patterns. Based on the segmentation results, a Random Forest model is developed to classify customers into churn and non-churn groups and identify high-risk customers. Experimental results show that the clustering model achieves a silhouette coefficient of 0.8797, while the Random Forest model outperforms other models with an accuracy of 0.825. The results demonstrate that the proposed framework effectively identifies high-value customers with elevated churn risk and supports more informed customer management decisions.

Biocompatibility of a biosensor can be achieved by grafting polymer brushes onto a solid surface. These brushes must be able to attract specific analytes or repel unwanted entities. This is obtained with weak polyelectrolyte polymer brushes that shrink or swell depending on external stimuli. In this study, the conformation of polyacrylic acid (PAA) and polyethylene oxide (PEO) polymer brushes was characterized as a function of pH and ionic strength using Atomic Force Microscopy (AFM) in spectroscopic mode. Instead of colloidal tips classically used to measure the mechanical behavior of the brush, force curves were performed with conventional tips for better sensitivity to the interaction between ions and polymer, which is responsible for their conformation. Since force mapping experiments generate thousands of curves, a statistical representation was employed to define the general trend of the curves and facilitate their interpretation. As expected, the neutral PEO is not affected by changes in solution pH and salinity. In contrast, PAA exhibits behaviors depending on the ions present in the solution and increasing salinity; the brush shrinks at low pH with H3O+ ions and swells with the addition of Na+ and K+ ions. The originality of the study also lies in the implementation of an Artificial Intelligence (AI) clustering model applied to force curves to specifically study a 50% PAA/50% PEO mixed polymer brush. This AI model makes it possible to distinguish areas of the surface where only one type of polymer has been grafted and to identify its nature according to its force curve.

Skeletal Class III (SCIII) malocclusion represents a heterogeneous cluster of craniofacial anomalies characterised by a sagittal mesial discrepancy. It is among the most challenging orthodontic conditions to treat because there is no standardisation regarding subphenotype classification or treatment efficacy prediction. This study aimed to develop a data-driven model to identify novel clinically relevant SCIII subphenotypes, contributing to tailored treatment protocols. A clinical subphenotypic classification model for SCIII was developed using 12 annotated craniofacial landmarks from lateral cephalometric radiographs of 655 adult SCIII patients of white origin. SCIII subphenotypes were identified by applying generalised Procrustes analysis and unsupervised clustering, and a classification model was developed for predicting subphenotypes for new patients. Cross-validation was employed to demonstrate the robustness of our clustering and classification models. Here we show that our model inferred six distinct subphenotypes that unravelled relevant morphological features in SCIII patients. We further demonstrate the generalisability of our model across ethnicities using an external validation cohort of patients of Korean origin. The identified SCIII subphenotypes exhibit a strong correlation with treatment decision. Our results contribute to the development of an accurate SCIII diagnostic tool (available at https://tools.istars.pt/sciii/), moving towards the goal of improving treatment efficacy for this condition.

Traditional rural experience group marketing has high cost, low efficiency, and insufficient accuracy. This paper combines e-commerce logs, social public opinion, and geographical indication data to build a closed loop of “feature engineering-clustering-segmentation-marketing”, aiming at reaching high-potential consumers in rural revitalization. Experimental results show that compared with K-means, the contour coefficient of the optimal clustering model is increased by 0.14 and the CH index is increased by 36%. In the A/B test, the conversion rate and customer unit price increased by 21.3% and 15.8% respectively, and the return rate decreased by 4.6%. Brand search volume increased by 42% and operating cost decreased by 12% in the first three months. Rolling retraining can control the performance fluctuation within 5% and explore the collaborative path of federated learning and privacy computing to provide reproducible digital value-added solutions for regional public brands.

ABSTRACT With the rapid development of maritime transportation, ship trajectory prediction has become critical for enhancing maritime safety. This study proposes Sea-KSformer, a novel ship trajectory prediction model that integrates a spatiotemporal sensing Transformer and a dynamic clustering mechanism. The model adopts multi-modal embedding, positional encoding, and causal self-attention to capture trajectory spatiotemporal dependencies while avoiding future information leakage. A partition modeling strategy is used to handle the inhomogeneity of AIS data. A fuzzy function improves robustness, and DTW-Entropy with density-driven initialization enhances K-Shape clustering. A memory enhancement mechanism further optimizes clustering. Experiments show that Sea-KSformer achieves superior performance in Haversine error, MAE, RMSE, noise resistance, and long-term prediction accuracy over existing methods.

High-dimensional data clustering poses significant challenges due to the curse of dimensionality, noise, and sparsity. Traditional clustering algorithms often struggle with scalability and accuracy in such contexts. To address these issues, this paper proposes a hybrid clustering model that integrates Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and Gaussian Mixture Models (GMM), leveraging the strengths of both approaches using machine learning techniques. DBSCAN efficiently identifies dense regions and eliminates noise, while GMM provides probabilistic soft clustering suitable for overlapping data distributions. By first using DBSCAN for pre-processing and noise reduction, followed by GMM for refined clustering, the hybrid model enhances performance in complex high-dimensional datasets. Dimensionality reduction techniques such as PCA and t-SNE are also incorporated to visualize and improve cluster quality. The proposed method is evaluated on benchmark high-dimensional datasets and compared against standalone clustering algorithms. Results demonstrate improved cluster compactness, separation, and computational efficiency, showcasing the effectiveness of the hybrid approach for high-dimensional data analysis in fields such as bioinformatics, image processing, and text mining.

Single-cell RNA sequencing (scRNA-seq) enables whole-transcriptomic profiling at single-cell resolution, facilitating the construction of virtual cell representations that capture the full spectrum of cellular identities. Realizing this goal hinges on accurate clustering, which remains challenging due to data sparsity, high dimensionality, noise, and the need to specify cluster numbers a priori. We propose scAClc, a novel clustering framework featuring multiobjective optimization and adaptive resolution discovery, designed to address these limitations through three key innovations. First, a Hierarchical Gene Relevance Module integrates global gene variability with local neighborhood-specific signals to eliminate redundancy while retaining biologically informative features. Second, an Anchor-Centered Contrastive Learning Module adaptively selects representative anchors to guide embedding learning, promoting compact intracluster structure and clear intercluster separation. Third, based on the robust low-dimensional embedding, we propose a Self-Adaptive Resolution Discovery Module to automatically infer the number of clusters by jointly modeling intra- and intercluster distances. Extensive experiments on 15 real scRNA-seq data sets demonstrate that scAClc consistently outperforms six state-of-the-art methods across multiple evaluation metrics. Ablation studies further confirmed the complementary contributions of each module. In addition, interpretability analysis effectively mitigates the "black box" nature of clustering models and sheds light on the biological mechanisms underlying cell clustering. The source code is publicly available at https://github.com/scAClc/scAClc.

Compared with previous generations, the 5G enhanced mobile broadband (eMBB) application delivers higher connection, quicker data speeds, and better customer support. Improving data transmission speeds for 5G uplink user equipment (UE) users is the goal of this study. Python is used for data analysis and framework building. This research looks at a 250-m-radius Picocell Base Station (PBS) that can have 15 user equipment (UEs). The position of the user is determined by the cell-range Poisson distribution. The physical base station (PBS), which assesses the state of the signal transmission channel, receives channel state information (CSI) from user equipment (UE). Rayleigh, Rician, free space path, and long-distance route loss models are used in the study. A dataset of channel statuses is generated by the query. There is dynamism in the dataset. K-means clustering is used by UEs to handle service-specific needs. By integrating bandwidth, clustering improves system performance and maximizes the cumulative rate of all user equipment. Channel gain, transmission rate, and minimum service information rate are the characteristics that define UEs. After grouping, users in Cluster 3 had the highest cumulative rate of 9.52 Mbps and an average rate of 7.52 Mbps. In addition to increasing system capacity, bandwidth concatenation satisfied the service needs for every user's equipment (UE). Performance criteria of several clustering models were evaluated, and K-means was found to be the best method. The method was methodically created to satisfy the goals of the study. This research investigates beamforming capabilities and adaptive clustering to improve user fairness and efficiency.

Social media sites such as Twitter (X) changed into high-velocity observatories of collective mood with millions of short, informal utterances recording public reactions to events, products, policies, and cultural moments in near real time. Mining the stream for actionable insight calls for approaches that respect two notoriously hard to handle properties of social text: (i) ambiguity — in reality, most posts display mixed or low-intensity affect rather than a single discrete label; and (ii) contextual drift — lexical and topical signals co-evolve with communities and time. Many standard sentiment pipelines requiring each message to be labeled as a single discrete class (positive/negative/neutral) fail to provide adequate behavioral insights for crisis monitoring or policy assessment. We close this gap by presenting a Sentiment-Aware Fuzzy Clustering model that models sentiment as a continuous signal and community mood as overlapping regions instead of disjoint boxes. We assign a polarity score to each post and then discretize the space using Fuzzy C-Means (FCM) to assign partial memberships to a number of emotional groups. This uncertainty-aware representation is more reflective of actual online behavior (i.e. a post could be strongly positive but still exhibit features of the neutral discourse) and serves as a principled basis for downstream interpretation in population scale.

Abstract Fleet-wide condition monitoring is a proven method for continuously assessing the operational health of an entire fleet, ensuring efficient and reliable performance. With the emergence of fleet-based monitoring, incorporating digital modeling, advanced diagnostics and predictive maintenance have become possible. This approach compares physical machines to their digital counterparts, enabling more sophisticated fault detection. A critical challenge addressed by fleet monitoring is managing large data volumes efficiently, avoiding the need for excessive high-frequency data. Our approach leverages SCADA data from similar wind turbines in the same region, assuming healthy operation for most turbines. Deviations in multiple measurements from normal conditions serve as fault indicators, enabling early detection and targeted interventions while reducing data processing demands. This study proposes an unsupervised learning method using statistical analysis of SCADA data, applied on a fleet of 20 wind turbines. After preprocessing to understand stochastic behaviors, a multi-feature hierarchical clustering (MFHC) model identifies patterns and groups turbines based on operational characteristics. By analyzing extracted features, the model efficiently detects faults and optimizes performance without requiring labeled datasets.

Mixture cure models are widely used in survival analysis to represent populations comprising both cured and susceptible subgroups. Moving beyond conventional approaches that address isolated challenges, this study introduces an integrated systems framework for analyzing complex survival data. The proposed model simultaneously accounts for three interdependent components: (1) the existence of a cured fraction, (2) the presence of incurable competing risks, and (3) within-cluster correlations. To address these issues, we propose a novel clustered survival model that simultaneously accommodates both cure fractions and competing risks. Our approach incorporates a working correlation matrix within estimating equations to model dependence structures - both in cure probabilities among cluster members and in survival times of susceptible individuals. The estimation procedure utilizes an ES algorithm framework for efficient computation of regression parameters. We conduct comprehensive simulation studies to evaluate the model’s finite-sample performance, demonstrating satisfactory results in parameter estimation and inference. Finally, the practical utility of our methodology is illustrated through an application to clinical trial data, providing empirical evidence of its effectiveness in real-world scenarios.

Assessment of pediatric emergencies by telephone is challenging and may be associated with uncertainty and overtriage. Video streaming has been introduced in emergency medical dispatch, but evidence regarding its impact in pediatric emergency calls remains limited. We evaluated associations between video-assisted dispatch assessment and urgency allocation, EMS resource use, and patient safety in pediatric calls. This predefined substudy of the cluster-randomized CAM-VISION trial included emergency calls concerning children ≤ 15years handled at the Emergency Medical Dispatch Center in the Central Denmark Region between January 1 and April 30, 2023. Dispatchers were randomized to video-assisted or telephone-only communication. Prespecified pediatric outcomes were the proportion of children assigned the lowest urgency level (response E), hospital admission within 24h after response E, and the distribution of all urgency levels (A-E). Additional analyses included EMS resource allocation, urgency changes between dispatch and scene, non-conveyance, time intervals, ICU admission, and 30-day mortality. Analyses followed an intention-to-treat approach using clustered regression models. Among 1,303 pediatric emergency calls, 586 were allocated to video-assisted dispatch and 717 to telephone-only communication. Video was successfully established in 74.7% of calls in the video group. Among predefined outcomes, the lowest urgency level (response E) was assigned more frequently in the video-assisted group (34.8% vs 28.0%; absolute difference 6.8 percentage points, 95% CI 0.1 to 13.4). No hospital admissions within 24h occurred among children dispatched at response E in either group. The highest urgency level (response A) was less frequent in the video-assisted group (37.9% vs 45.0%; absolute difference - 7.2 percentage points, 95% CI - 14.0 to - 0.3). Additional analyses showed that physician-staffed vehicles arrived less often in the video-assisted group (35.3% vs 44.2%; absolute difference - 8.9 percentage points, 95% CI - 16.8 to - 0.9). Median time from call to dispatch was one minute longer in the video-assisted group, while on-scene time and hospital length of stay were similar between groups. Video-assisted dispatch in pediatric emergency calls was associated with more frequent assignment of the lowest urgency level and reduced use of physician-staffed vehicles without evidence of compromised patient safety. ClinicalTrials.gov identifier NCT05742412.

A substantial proportion of ischemic strokes remain classified as embolic stroke of undetermined source (ESUS) despite standard diagnostic evaluation. Prolonged cardiac monitoring with implantable cardiac monitors (ICMs) increases atrial fibrillation (AF) detection, but the optimal timing of ICM implantation after ESUS remains uncertain. To evaluate whether early versus delayed ICM implantation after ESUS influences AF detection and time to diagnosis. We conducted a multicenter observational cohort study emulating a target trial. Consecutive ESUS patients undergoing ICM implantation were classified as ICMEARLY (⩽30 days) or ICMDELAYED (31-365 days) implantation after the index event. Inverse probability weighting was applied to adjust for baseline confounding. Primary and secondary outcomes included AF detection within 30, 90, and 120 days after implantation, assessed using weighted logistic regression, Poisson models for detection rates per person-time, Cox proportional hazards models, and restricted mean survival time (RMST). Sensitivity analyses included center-level clustering and competing-risk models. Among 333 patients (90 ICMEARLY and 243 ICMDELAYED), early implantation was associated with significantly higher AF detection within 30 days (7.8% vs 1.6%; odds ratio (OR) = 4.49, 95% confidence interval (CI) = 1.17-17.27; p = 0.028) and higher detection rates per person-time (incidence rate ratio (IRR) = 4.26, 95% CI = 1.16-15.60; p = 0.029). Consistent associations were observed at 90 and 120 days. Time-to-event analyses showed higher hazards of AF detection with early implantation (hazard ratio (HR) = 4.29 at 30 days; HR = 2.97 at 90 days; HR = 2.77 at 120 days; all p < 0.01). RMST analyses demonstrated progressively shorter time to AF diagnosis in the ICMEARLY group across multiple time horizons. Results were robust across sensitivity analyses. Early ICM implantation after ESUS is associated with higher and faster AF detection compared with delayed implantation. When ICM monitoring is indicated, avoiding unnecessary delays may substantially enhance diagnostic yield.