Search Results (16)

Deep learning models for brain tumor classification from magnetic resonance imaging (MRI) often achieve high in-dataset accuracy but exhibit substantial performance degradation when evaluated on unseen clinical data due to domain shift arising from variations in imaging protocols and intensity distributions. Existing approaches largely rely on architectural scaling or parameter-level regularization, which do not explicitly constrain the stability of learned feature representations. This manuscript proposes Feature Space Stability Regularization (FSSR), a lightweight and model-agnostic training framework that enforces consistency in latent feature representations under realistic, MRI-safe-intensity perturbations. FSSR introduces an auxiliary feature space loss that minimizes the ${\ell }_2$ distance between normalized embeddings extracted from the input MRI images and their intensity-perturbed counterparts, alongside standard cross-entropy supervision. This manuscript evaluated FSSR across three convolutional backbones, ResNet-18, ResNet-34, and DenseNet-121, trained exclusively on the Kaggle Brain MRI dataset. Feature space analysis demonstrates that FSSR consistently reduces mean feature deviation and variance across architectures, indicating more stable internal representations. Generalization is assessed via zero-shot evaluation on the fully unseen BRISC-2025 dataset without retraining or fine-tuning. On the source domain, the best-performing configuration achieves 97.71% accuracy and 97.55% macro-F1. Under domain shift, FSSR improves external accuracy by up to 8.20 percentage points and the macro-F1 by up to 12.50 percentage points, with DenseNet-121 achieving a 96.70% accuracy and 96.87% macro-F1 at a domain gap of only 0.94%. Confusion matrix analysis further reveals the reduced class confusion and more stable recall across challenging tumor categories, demonstrating that feature-level stability is a key factor for robust brain MRI classification under domain shift. Full article

Deep learning approaches for egocentric video understanding often lack a principled theoretical treatment of stability, particularly when dealing with the sparse, noisy, and temporally ambiguous observations characteristic of first-person imaging. In this work, we frame egocentric video question answering not merely as a classification task, but as an ill-posed inverse problem aimed at reconstructing latent semantic intent from stochastically perturbed visual signals. To address the instability inherent in standard dual-encoder architectures, we present a framework with a mathematical interpretation that incorporates gated cross-modal interaction within the transformer backbone. Formally, the video-side update analyzed in this work is defined as a learnable convex combination of unimodal feature representations and cross-modal attention residuals; the full implementation applies analogous gated cross-modal updates bidirectionally. From a regularization perspective, the gating mechanism can be interpreted as an adaptive parameter that balances data fidelity against language-conditioned structural constraints during feature reconstruction. We provide the Bounded Update Property (Lemma 1) and an analytical layer-wise sensitivity bound and empirically demonstrate that the proposed framework achieves measurable improvements in both accuracy and stability on the EgoTaskQA and MSR-VTT benchmarks. On EgoTaskQA, our model improves accuracy from 27.0% to 31.7% (+4.7 pp) and reduces the accuracy drop under 50% frame drop from 3.93 pp to 0.94 pp. On MSR-VTT, our model improves accuracy by 13.0 pp over the dual-encoder baseline. Under severe perturbation (50% frame drop) on MSR-VTT, our model retains 97.7% of its clean performance, whereas the baseline exhibits near-zero drop accompanied by majority-class behavior. These results provide empirical evidence that the proposed interaction induces stable behavior under perturbations in an ill-posed multimodal inference setting, mitigating sensitivity to sampling variability while preserving query-relevant temporal structure. Furthermore, an entropy-based analysis indicates that the gating mechanism prevents excessive diffusion of attention, promoting coherent temporal reasoning. Overall, this work offers a mathematically informed perspective on designing interaction mechanisms for stable multimodal systems, with a focus on robust reasoning under temporal ambiguity. Full article

Accurate and early detection of bone cancer is critical for improving patient outcomes, yet conventional radiographic interpretation remains limited by subjectivity and variability. Conventional AI models often struggle with complex multi-modal noise distributions, non-convex and topologically entangled latent manifolds, extreme class imbalance in rare oncological conditions, and heterogeneous data fusion constraints. To address these challenges, we present a Quantum-Inspired Classical Convolutional Neural Network (QC-CNN) inspired by quantum analogies for automated bone cancer detection in radiographic images. The proposed architecture integrates classical convolutional layers for hierarchical feature extraction with a classical variational layer motivated by high-dimensional Hilbert space analogies for enhanced pattern discrimination. A curated and annotated dataset of bone X-ray images was utilized, partitioned into training, validation, and independent test cohorts. The QC-CNN was optimized using stochastic gradient descent (SGD) with adaptive learning rate scheduling, and regularization strategies were applied to mitigate overfitting. Quantitative evaluation demonstrated superior diagnostic performance, achieving high accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). Results highlight the ability of classical CNN with quantum-inspired design to capture non-linear correlations and subtle radiographic biomarkers that classical CNNs may overlook. This study establishes QC-CNN as a promising framework for quantum-analogy motivated medical image analysis, providing evidence of its utility in oncology and underscoring its potential for translation into clinical decision-support systems for early bone cancer diagnosis. All computations in the present study are performed using classical algorithms, with quantum-inspired concepts serving as a conceptual framework for model design and motivating future extensions. Full article

Background/Objectives: In Japan, the number of elementary and junior high school students who do not attend school is increasing. Sleep problems are considered a contributing factor. Methods: This study utilized self-administered questionnaires about the sleep patterns and backgrounds of 25,257 students from the 3rd–10th grades across 91 elementary schools, 51 junior high schools, and 36 high schools in Japan. Latent class analysis was performed to assess sleep regularity. Logistic regression analysis was conducted to examine the relationship between sleep regularity and school attendance status, as well as the relationship with protective factors against non-attendance. Results: Overall, 19,005 students responded. The response rate was 75.2%. Sleep regularity was categorized into Class 1, Regular; Class 2, Somewhat Irregular; Class 3, Irregular; and Class 4, Schedule-Dependent. Class 1 decreased with grade, from 61.8% in the 3rd grade to 46.2% in the 10th grade. Class 3 comprised 10.0% of students not experiencing school non-attendance, 37.9% among students with persistent school non-attendance, and 17.9% among students who had resumed school attendance after school non-attendance in the previous year. Classes 2, 3, and 4 showed a negative relationship with protective factors against non-attendance such as good relationships with teachers and family, good communication, academic performance, proficiency in athletic activities, and the presence of a place to belong outside school. Conclusions: Sleep irregularity is related to school non-attendance and may serve as a barometer of students’ communication and academic difficulties. Additionally, we propose an early intervention for sleep problems to prevent the exacerbation of school non-attendance. Full article

Electroencephalography (EEG)-based motor imagery (MI) brain–computer interfaces (BCIs) hold considerable potential for applications in neuro-rehabilitation and assistive technologies. Yet, their development remains constrained by challenges such as low spatial resolution, vulnerability to noise and artifacts, and pronounced inter-subject variability. Conventional approaches, including common spatial patterns (CSP) and convolutional neural networks (CNNs), often exhibit limited robustness, weak generalization, and reduced interpretability. To overcome these limitations, we introduce EEG-GCIRNet, a Gaussian connectivity-driven EEG imaging representation network coupled with a regularized LeNet architecture for MI classification. Our method integrates raw EEG signals with topographic maps derived from functional connectivity into a unified variational autoencoder framework. The network is trained with a multi-objective loss that jointly optimizes reconstruction fidelity, classification accuracy, and latent space regularization. The model’s interpretability is enhanced through its variational autoencoder design, allowing for qualitative validation of its learned representations. Experimental evaluations demonstrate that EEG-GCIRNet outperforms state-of-the-art methods, achieving the highest average accuracy (81.82%) and lowest variability (±10.15) in binary classification. Most notably, it effectively mitigates BCI illiteracy by completely eliminating the “Bad” performance group (<60% accuracy), yielding substantial gains of ∼22% for these challenging users. Furthermore, the framework demonstrates good scalability in complex 5-class scenarios, performing competitive classification accuracy (75.20% ± 4.63) with notable statistical superiority (p = 0.002) against advanced baselines. Extensive interpretability analyses, including analysis of the reconstructed connectivity maps, latent space visualizations, Grad-CAM++ and functional connectivity patterns, confirm that the model captures genuine neurophysiological mechanisms, correctly identifying integrated fronto-centro-parietal networks in high performers and compensatory midline circuits in mid-performers. These findings suggest that EEG-GCIRNet provides a robust and interpretable end-to-end framework for EEG-based BCIs, advancing the development of reliable neurotechnology for rehabilitation and assistive applications. Full article

Time-series anomaly detection is imperative for ensuring reliability and safety in intelligent manufacturing systems. However, real-world environments typically provide only normal operating data and exhibit significant periodicity, noise, imbalance, and domain variability. The present study proposes CL-OCC, a contrastive learning-based one-class framework that integrates seasonal-trend decomposition using loess (STL) for structure-preserving temporal augmentation, a cosine-regularized soft boundary for compact normal-region formation, and variance-preserving regularization to prevent latent collapse. A convolutional recurrent encoder is first pretrained via an autoencoder objective and subsequently optimized through a unified loss that balances contrastive invariance, soft-boundary constraint, and variance dispersion. Experiments on semiconductor equipment data and three public benchmarks demonstrate that CL-OCC provides competitive or superior performance relative to reconstruction-, prediction-, and contrastive-based baselines. CL-OCC exhibits smoother anomaly trajectories, earlier detection of gradual drifts, and strong robustness to noise, window-length variation, and extreme class imbalance. A study of the effects of ablation and interaction on the stability of representations indicates that STL-based augmentation, boundary shaping, and variance regularization contribute complementary benefits to this stability. While the qualitative results indicate limited sensitivity to extremely short impulsive disturbances, the proposed framework delivers a generalizable and stable solution for unsupervised industrial monitoring, with promising potential for extension to multi-resolution analysis and online prognostics and health management (PHM) applications. Full article

Introduction: Regular physical activity (PA) has a beneficial effect on joint pain, stiffness, strength, flexibility, and aerobic capacity in patients with rheumatoid arthritis (RA). Objective: The aim of this study was to assess the level of PA in patients with rheumatoid arthritis and to identify potential barriers to this activity. Material and Methods: The study involved 132 patients with RA. Participants completed the International Physical Activity Questionnaire (IPAQ), the Functional Assessment of Chronic Illness Therapy—Fatigue Scale (FACIT-F), the Tampa Scale for kinesiophobia (TSK), Strength, Ambulation, Rising from a chair, Stair climbing and history of Falling questionnaire (SARC-F) for sarcopenia assessment, and the Patient Health Questionnaire-9 (PHQ-9) for depression. Basic socio-epidemiological data, disease activity score in 28 joints (DAS₂₈), duration of disease, and therapy information were retrieved from electronic patient records. Latent class analysis (LCA) was used to identify subpopulations of patients. Results: The study included 109 women (82.6%) and 23 men (17.4%). Low levels of PA were observed in 16 patients (12%), moderate levels in 70 patients (53%), and high levels in 42 patients (35%). Symptoms of pronounced fatigue were significantly associated with low PA (28.5 ± 11.3 vs. 37 ± 7 vs. 37 ± 10; p = 0.002). The risk of sarcopenia was significantly higher in RA patients with low PA (p = 0.05). Kinesiophobia was present in all three groups (65.2%). LCA identified two classes. In the first class, patients were more likely to be non-exercisers compared to the second class. Patients in the first class were characterized by a higher probability of being female, obese, with lower education levels. Patients in the first class had pronounced fatigue, kinesiophobia and more frequent symptoms of depression. The second class (65% of the total population) included patients who exercised moderately to frequently (93%) and were middle-aged. They were less obese, highly educated, employed, and majority of them achieved low disease activity or remission. In addition, they had lower risks for sarcopenia, depression, fatigue, and kinesiophobia. Conclusions: This study showed that RA patients with moderate and high levels of PA have better disease control, fewer symptoms of fatigue and depression, and a lower risk of sarcopenia. However, kinesiophobia was significantly present in all three groups, indicating a need for further promotion of this non-pharmacological treatment. Full article

The causes and conditions of displacement often increase the vulnerability of migrant and refugee populations to food insecurity, alongside other material hardships. We aimed to examine the multidimensional aspects and patterns of food insecurity and other material hardships in a cross-sectional sample of 6221 Venezuelan refugees and migrants in urban Colombia using a latent class analysis. Using multinomial and logistic regression models, we investigated the demographic and migratory experiences associated with identified classes and how class membership is associated with multiple health outcomes among Venezuelan refugees and migrants, respectively. Approximately two thirds of the sample was comprised cisgender women, and the participants had a median age of 32 years (IQR: 26–41). Four heterogeneous classes of food insecurity and material hardships emerged: Class 1—low food insecurity and material hardship; Class 2—high food insecurity and material hardship; Class 3—high income hardship with insufficient food intake; and Class 4—income hardship with food affordability challenges. Class 2 reflected the most severe food insecurity and material hardships and had the highest class membership; Venezuelans with an irregular migration status were almost 1.5 times more likely to belong to this class. Food insecurity and material hardship class membership was independently associated with self-rated health, mental health symptoms, and recent violence victimization and marginally associated with infectious disease outcomes (laboratory-confirmed HIV and/or syphilis infection). Social safety nets, social protection, and other interventions that reduce and prevent material hardships and food insecurity among refugees and migrants, alongside the host community, may improve public health, support development, and reduce healthcare costs. In the long term, regularization and social policies for migrants aimed at enhancing refugees’ and migrants’ social and economic inclusion may contribute to improving food security in this population. Full article

This paper investigates the latent classes of parking preference for drivers and the economic effects after implementing Parking Space Comprehensive Utilization (PSC) in traditional business districts (TBD), with a particular focus on the parking preferences of electric vehicle users (EVU). Firstly, Exploratory Factor Analysis (EFA) is used to reduce dimensionality and determine the latent structure. Then, based on the Latent Class Model (LCM), the customers are classified, and the proportion of each class under various latent variables is analyzed. Finally, the paper conducts a quantitative analysis of economic effects by considering different psychological factors across different customer classes. With the data obtained from revealed preference (RP) and stated preference (SP) surveys, this paper identifies the customers’ preferences for the three scenarios presented. The results show that (1) customers can be classified into four classes: core customers (CCS, 34%), potential customers (PCS, 29%), regular customers (RCS, 22%), and marginal customers (MCS, 15%), among which EVU do not show a significant preference for parking charging facilities in TBD; (2) the potential economic improvements for these four classes are: 9%, 12%, 8%, and 10%; (3) CCS has the greatest potential to increase store revenue by ¥7041, while PCS has the greatest potential to increase store customer flow by 31%. These findings provide a valuable reference for decision-making by TBD store managers. Full article

The retrospective reporting of users’ hearing aid (HA) usage can provide insight into individualized HA usage patterns. Understanding these HA usage patterns can help to provide a tailored solution to meet the usage needs of HA users. This study aims to understand the HA usage pattern in daily-life situations from self-reported data and to examine its relationship to self-reported outcomes. A total of 1537 participants who responded to questions related to situations where they always took off or put on the HAs were included in the study. A latent class analysis was performed to stratify the HA users according to their HA usage pattern. The results showed distinct usage patterns in the latent classes derived for both scenarios. The demographics, socio-economic indicators, hearing loss, and user-related factors were found to impact HA usage. The results showed that the HA users who reported using the HAs all the time (regular users) had better self-reported HA outcomes than situational users, situational non-users, and non-users. The study explained the underlying distinct HA usage pattern from self-reported questionnaires using latent class analysis. The results emphasized the importance of regular use of HAs for a better self-reported HA outcome. Full article

Reconstruction-based approaches to anomaly detection tend to fall short when applied to complex datasets with target classes that possess high inter-class variance. Similar to the idea of self-taught learning used in transfer learning, many domains are rich with similar unlabeled datasets that could be leveraged as a proxy for out-of-distribution samples. In this paper we introduce the latent-insensitive autoencoder (LIS-AE) where unlabeled data from a similar domain are utilized as negative examples to shape the latent layer (bottleneck) of a regular autoencoder such that it is only capable of reconstructing one task. We provide theoretical justification for the proposed training process and loss functions along with an extensive ablation study highlighting important aspects of our model. We test our model in multiple anomaly detection settings presenting quantitative and qualitative analysis showcasing the significant performance improvement of our model for anomaly detection tasks. Full article

This paper investigates preference heterogeneity among Slovenian energy consumers and attempts to ascertain how different consumer groups value various attributes of energy products and services. More specifically, it aims to establish whether a consumer segment can be identified that shows a preference for additional energy services—in particular services, associated with energy-efficient and green behaviour. A latent class analysis is employed to classify consumers on the basis of their preferences for energy services. Additionally, information about their attitudes and behaviour toward green energy and energy efficiency, energy consumption, and usage of energy services together with socio-economic characteristics is used in the latent class regression to explain differences between latent consumer classes. Three classes are identified: the largest class of regular consumers, energy-efficient consumers, and dissatisfied consumers. In contrast to regular and dissatisfied consumers, energy-efficient consumers show a significantly higher interest in additional services, energy efficiency, and green energy. In line with the found heterogeneity of consumer preferences, suppliers should customise marketing strategies to meet the needs of specific segments. Energy policymakers also need to pay more attention to consumer heterogeneity and behavioural changes to increase the effectiveness of energy efficiency policies. Full article

The last series of Raven’s standard progressive matrices (SPM-LS) test was studied with respect to its psychometric properties in a series of recent papers. In this paper, the SPM-LS dataset is analyzed with regularized latent class models (RLCMs). For dichotomous item response data, an alternative estimation approach based on fused regularization for RLCMs is proposed. For polytomous item responses, different alternative fused regularization penalties are presented. The usefulness of the proposed methods is demonstrated in a simulated data illustration and for the SPM-LS dataset. For the SPM-LS dataset, it turned out the regularized latent class model resulted in five partially ordered latent classes. In total, three out of five latent classes are ordered for all items. For the remaining two classes, violations for two and three items were found, respectively, which can be interpreted as a kind of latent differential item functioning. Full article

This study aims at combining time and quantity of carbohydrate (CH) intake in the definition of eating patterns in UK adults and investigating the association of the derived patterns with type 2 diabetes (T2D). The National Diet and Nutrition Survey (NDNS) Rolling Program included 6155 adults in the UK. Time of the day was categorized into 7 pre-defined time slots: 6–9 am, 9–12 noon, 12–2 pm, 2–5 pm, 5–8 pm, 8–10 pm, and 10 pm–6 am. Responses for CH intake were categorized into: no energy intake, CH <50% or ≥50% of total energy. Non-parametric multilevel latent class analysis (MLCA) was applied to identify eating patterns of CH consumption across day-time, as a novel method accounting for the repeated measurements of intake over 3–4 days nested within individuals. Survey-designed multivariable regression was used to assess the associations of CH eating patterns with T2D. Three CH eating day patterns (low, high CH percentage and regular meal CH intake day) emerged from 24,483 observation days; based on which three classes of CH eaters were identified and characterized as: low (28.1%), moderate (28.8%) and high (43.1%) CH eaters. On average, low-CH eaters consumed the highest amount of total energy intake (7985.8 kJ) and had higher percentages of energy contributed by fat and alcohol, especially after 8 pm. Moderate-CH eaters consumed the lowest amount of total energy (7341.8 kJ) while they tended to have their meals later in the day. High-CH eaters consumed most of their carbohydrates and energy earlier in the day and within the time slots of 6–9 am, 12–2 p.m. and 5–8 pm, which correspond to traditional mealtimes. The high-CH eaters profile had the highest daily intake of CH and fiber and the lowest intake of protein and fat. Low-CH eaters had greater odds than high-CH eaters of having T2D in self-reported but not in previously undiagnosed diabetics. Further research using prospective longitudinal studies is warranted to ascertain the direction of causality in the association of CH patterns with type 2 diabetes. Full article

Due to the constant increase in Earth Observation (EO) data collections, the monitoring of land cover is facilitated by the temporal diversity of the satellite images datasets. Due to the capacity of Synthetic Aperture Radar (SAR) sensors to operate independently of sunlight and weather conditions, SAR image time series offer the possibility to form a dataset with almost regular temporal sampling. This paper aims at mining the SAR image time series for an analysis of target’s behavior from the perspective of both temporal evolution and coherence. The authors present a two-level analytical approach envisaging the assessment of global (related to perceivable structures on the ground) and local (related to changes occurred within a perceivable structure on the ground) evolution inside the scene. The Latent Dirichlet Allocation (LDA) model is implemented to identify the categories of evolution present in the analyzed scene, while the statistical and coherent proprieties of the dataset’s images are exploited in order to identify the structures with stable electromagnetic response, the so-called Persistent Scatterers (PS). A comparative study of the two algorithms’ classification results is conducted on ERS and Sentinel-1 data. At global scale, the results fit human perception, as most of the points which can be exploited for Persistent Scatterers Interferometry (PS-InSAR) are classified within the same class, referring to stable structures. At local scale, the LDA classification demands for an extended number of classes (defined through a perplexity-based analysis), enabling further differentiation inside the evolutional character of those stable structures. The comparison against the map of detected PS reveals which classes present higher temporal correlation, determining a stable evolutionary character, opening new perspectives for validation of both PS detection and SITS analysis algorithms. Full article