Reliably detecting COVID-19 is critical for diagnosis and disease control. However, imbalanced data in medical datasets pose significant challenges for machine learning models, leading to bias and poor generalization. The dataset obtained from the EPIVIGILA system and the Chilean Epidemiological Surveillance Process contains information on over 6,000,000 patients, but, like many current datasets, it suffers from class imbalance. To address this issue, we applied various machine learning algorithms, both with and without sampling methods, and compared them using different classification and diagnostic metrics such as precision, sensitivity, specificity, likelihood ratio positive, and diagnostic odds ratio. Our results showed that applying sampling methods to this dataset improved the metric values and contributed to models with better generalization. Effectively managing imbalanced data is crucial for reliable COVID-19 diagnosis. This study enhances the understanding of how machine learning techniques can improve diagnostic reliability and contribute to better patient outcomes.

We consider the detection and characterization of brain state transitions based on ongoing electroencephalography (EEG). Here, a brain state represents a specific brain dynamical regime or mode of operation that produces a characteristic quasi-stable pattern of activity at the topography, sources, or network levels. These states and their transitions over time can reflect fundamental computational properties of the brain, shaping human behavior and brain function. The hidden Markov model (HMM) has emerged as a useful tool for uncovering the hidden dynamics of brain state transitions based on observed data. However, the limitations of the Geometric distribution of states’ durations (dwell times) implicit in the standard HMM, make it sub-optimal for modeling brain states in EEG. We propose using hidden semi Markov models (HSMM), a generalization of HMM that allows modeling the brain states duration distributions explicitly. We present a Bayesian formulation of HSMM and use the variational Bayes framework to efficiently estimate the HSMM parameters, the number of brain states, and select among candidate brain state duration distributions. We assess HSMM performance against HMM on simulated data and demonstrate that the accurate modeling of state durations is paramount for making reliable inference when the task at hand requires accurate model predictions. Finally, we use actual resting-state EEG data to illustrate the benefits of the approach in practice. We demonstrate that the possibility of modeling brain state durations explicitly provides a new way for investigating the nature of the neural dynamics that generated the EEG data.

The global rise in the aging population and the increase in older adults living alone have raised concerns about health-related behaviors, particularly sedentary lifestyles and reduced daily activities. These behaviors are linked to higher risks of physical and cognitive conditions. While many global studies have explored these patterns, research within the Chilean context remains limited. This work presents a analysis of behavioral patterns in elderly individuals living alone in Chile, offering valuable insights into this population. Using clustering techniques, we identified two distinct activity patterns among the participants. The first pattern is characterized by a gradual increase in activity during the day, peaking around midday and followed by a decline, likely associated with meal preparation and rest. The second pattern demonstrates a more dynamic lifestyle, with a rapid surge in activity after waking and sustained levels throughout the day, suggesting a potentially healthier approach to aging. These findings align with previous studies indicating high levels of sedentary behavior in older adults, reinforcing the need for interventions tailored to diversify daily routines and promote physical activity. This study is the first to explore these patterns in the Chilean context, contributing to a more comprehensive understanding of elderly care and informing future strategies for improving the well-being of older adults living alone.

Background: Class imbalance remains a major limitation in artificial intelligence based oral cancer diagnosis, particularly in small clinical image datasets where malignant lesions are underrepresented. Methods: Using a dataset of 3000 mobile-acquired oral cavity images categorized into four diagnostic classes (healthy, benign, oral potentially malignant disorders, and oral cancer), we systematically evaluated the impact of imbalance mitigation strategies on diagnostic performance. Three backbone architectures, EfficientNet, Vision Transformer, and Swin Transformer, were trained under four conditions: raw imbalanced training, random under-sampling, random over-sampling, and medical-safe data augmentation. Performance was evaluated using stratified five-fold cross-validation, and all reported metrics correspond to the mean performance across the validation folds. Results: In multiclass evaluation, augmentation yielded the strongest overall performance, with EfficientNet achieving the highest macro-${\mathrm{F}}_1$ score (0.669 ± 0.024) and area under curve (0.881 ± 0.015). Under clinically oriented binary malignant-risk evaluation (high-risk vs. low-risk lesions), sensitivity reached 0.817 ± 0.030 with EfficientNet, while data augmentation maintained competitive malignant-risk detection across the evaluated architectures. In contrast, random undersampling consistently yielded lower descriptive performance across architectures. Conclusions: These findings indicate that augmentation-based imbalance mitigation is associated with improved multiclass and oral cancer specific performance while maintaining competitive overall discrimination. The results provide practical guidance for developing clinically robust artificial intelligence systems for early oral cancer detection in imbalanced real-world datasets.

Falls among older adults represent a critical public health issue, being one of the leading causes of severe injuries and accident-related mortality. because of this, Fall Detection Systems face significant challenges due to the complexity of environments and variability in movement patterns. On this matter, Data fusion, including method-level, sensor-level, feature-level, and decision-level approaches, offers a promising method to enhance detection accuracy. However, decision-level fusion remains under-explored, despite its potential to integrate outputs from multiple classifiers to address these limitations. This article presents a comparative study on the performance of decision-level fusion strategies in fall detection systems, employing different configurations of sensors, classifiers, and decision fusion algorithms. A dataset was collected using four sensor modalities: Accelerometer, Infrared Camera, Light Detection and Ranging, and Radar. Three classifiers were employed, and five decision-level fusion strategies were evaluated. Performance was assessed across three levels of simulated noise using four metrics: accuracy, sensitivity, specificity, and precision. The implementation of decision-level fusion strategies resulted in improvements in the performance of fall detection systems. Accuracy increased from 85% without decision-level fusion to 93% using Random Forest. Among deterministic strategies, Voting with Accuracy achieved competitive results. Light Detection and Ranging, demonstrated robustness under noise, while other sensors showed varying sensitivity to noise. These results demonstrate the importance of utilizing decision-level fusion strategies, as they lead to improvements in system performance. This enhancement is achieved by leveraging the complementarity of multiple technologies, allowing systems to adapt effectively to diverse and dynamic environments. This functionality is particularly useful in standalone or closed solutions, where such complementarity can address inherent limitations and enhance overall system capabilities.

BACKGROUND Cancer research increasingly relies on heterogeneous datasets such as electronic health records (EHRs), genomic sequences, imaging, and multi-omics data. Robust research infrastructures are required to integrate these sources, yet existing approaches remain fragmented. Knowledge models, including ontologies, taxonomies, and knowledge graphs, provide standardized frameworks for data structuring, interoperability, and reuse, but their effectiveness in oncology research is not fully synthesized. OBJECTIVE This study aimed to systematically review knowledge models applied in cancer research infrastructures and to evaluate their role in addressing data heterogeneity, scalability, We conducted a systematic literature review following the guidelines of Kitchenham and Charters and the PRISMA 2020 framework. Searches were performed in PubMed, Scopus, Web of Science, and other digital libraries for studies published between 2015 and 2025. Inclusion criteria required that articles describe or evaluate knowledge models used in cancer data repositories. Three reviewers independently screened and assessed studies using predefined quality criteria. Data extraction covered model type, application domain, data types, methodological approaches, and reported challenges. We summarized the findings in a PRISMA flow diagram. and interoperability. METHODS We conducted a systematic literature review following the guidelines of Kitchenham and Charters and the PRISMA 2020 framework. Searches were performed in PubMed, Scopus, Web of Science, and other digital libraries for studies published between 2015 and 2025. Inclusion criteria required that articles describe or evaluate knowledge models used in cancer data repositories. Three reviewers independently screened and assessed studies using predefined quality criteria. Data extraction covered model type, application domain, data types, methodological approaches, and reported challenges. We summarized the findings in a PRISMA flow diagram. RESULTS A total of 39 primary studies were included. Most studies (over 75%) employed formal ontologies, particularly the National Cancer Institute Thesaurus (NCIT) and Systematized Nomenclature of Medicine-Clinical Terms (SNOMED-CT), to ensure semantic interoperability. Knowledge graphs, thesauri, and taxonomies were also reported. The models were applied to genomics, proteomics, clinical trials, imaging, and patient-reported outcomes. Key contributions included enhanced semantic data integration, ontology-based data structuring, and improved information retrieval and decision support. Reported challenges included scalability limitations in large-scale applications, manual curation demands, ontology evolution, and restricted usability for clinicians. Opportunities highlighted in the literature involved machine learning-assisted ontology enrichment, modular architecture, and community-driven standards such as the Open Biological and Biomedical Ontologies (OBO) Foundry. CONCLUSIONS Knowledge models are central to enabling interoperable and reusable infrastructures in cancer research. Ontologies and knowledge graphs provide essential foundations for semantic integration across heterogeneous biomedical data. Despite advances, challenges remain in scalability, sustainability, and real-world adoption within clinical workflows. Future research should prioritize automation, user-centered tools, and cross-institutional collaboration to ensure that semantic technologies can transition from prototypes to operational infrastructures supporting precision oncology. CLINICALTRIAL Not applicable - this study is a systematic review and did not involve a clinical trial.