Conference Agenda

This study assesses the amalgamation of High-Performance Computing (HPC) architectures with deep learning models for expedited brain MRI analysis in clinical diagnostic environments. We conducted a systematic comparison of three computing configurations available to Latin American universities: cloud-based CPU instances (Linode G8 Dedicated), cloud-based GPU instances (Linode RTX4000 Ada), and an on-premise workstation (Dell Precision 7960 Tower with dual RTX 5000 Ada GPUs). We utilized a dataset of 200 annotated brain magnetic resonance imaging (MRI) scans for binary classification of neurological abnormalities (presence/absence of lesions ≥5mm, including white matter hyperintensities, tumors, and vascular malformations) as determined by consensus of two board-certified neuroradiologists to train ResNet-50 and Vision Transformer models, assessing training efficiency, inference delay, energy consumption, and cost-effectiveness. The results indicate that the Dell Precision workstation attained an 11.1× acceleration in training duration (12.8 versus 142.7 minutes) relative to CPU-only cloud instances, while inference latency was minimized to 8.7 ms per image.

Traditional health monitoring systems rely on periodic sampling, which introduces significant delays in detecting critical physiological events. This paper presents an IoT-based biomedical architecture integrating real-time streaming analytics to enable early detection of adverse health conditions. The proposed system combines wearable biomedical sensors, MQTT communication protocols, and Apache Kafka-based streaming pipelines for continuous physiological data processing. We evaluated the architecture through controlled simulations comparing streaming analytics against conventional periodic sampling approaches as a theorical approach. Results demonstrate a substantial reduction in event detection latency, with the streaming system achieving mean detection delays of 47.0 seconds compared to 300.0 seconds for periodic sampling (p < 0.001). The MQTT-based communication layer exhibited mean latency of 30.1 ms with 95th percentile at 56.8 ms, significantly outperforming HTTP alternatives (mean: 149.0 ms, 95th percentile: 289.5 ms). System scalability testing revealed linear throughput scaling, supporting up to 100 concurrent sensors at 88,000 messages per second. These findings validate the efficacy of streaming analytics in biomedical IoT systems for preventive healthcare applications, with relevance for continuous monitoring of cardiac, respiratory, and metabolic parameters.

The shift toward hybrid education models in engineering programs has created the need to develop intelligent systems capable of personalizing learning experiences. This study presents a methodological framework for implementing learning analytics aimed at the early identification of students at academic risk and the design of personalized interventions in engineering faculties. A predictive model was developed based on data from learning management systems (LMS), formative assessments, and student activity patterns. The methodology included Monte Carlo simulation with 1,000 iterations for model validation, applied to a simulated cohort of 847 students over four academic semesters. The predictive model achieved an area under the ROC curve of 0.803 (95% CI: 0.744–0.845), with a sensitivity of 70.9% and a specificity of 77.7%. The personalized interventions implemented demonstrated statistically significant improvements in the academic performance of the treatment group (GPA: 2.68 ± 0.59) compared to the control group (GPA: 2.38 ± 0.64), with a medium effect size (Cohen’s d = 0.49, p < 0.001). The retention rate increased from 73.1% to 88.6% among students who received the intervention (χ² = 12.08, p < 0.001). The results suggest that integrating learning analytics into hybrid education environments constitutes an effective strategy for curriculum personalization and for reducing the risk of dropout in engineering programs.

Continuity of maternal care is a critical determinant of perinatal outcomes. In Honduras, despite advances in service coverage, gaps persist in the retention of women along the continuum of care. To model the transitions between prenatal care and institutional delivery states using Markov chains, to simulate scenarios of improvement in coverage and retention, and to estimate their impact on continuity of care indicators. Data from Honduras' ENDESA/MICS 2019 were used to parameterize a six-state Markov model representing the maternal care continuum. Stratified transition matrices were estimated by area of residence and wealth quintile, and four intervention scenarios were simulated using Monte Carlo simulation with 10,000 iterations. The probability of achieving institutional delivery was 95.0% (95%CI: 66.2-96.0%) at the national level, with significant disparities between urban (97.1%) and rural (91.7%) areas, and between the richest (97.1%) and poorest (89.5%) quintiles. The main bottleneck identified was the transition from partial prenatal care to institutional delivery in rural populations. The simulations indicated that interventions focused on improving access to institutional childbirth (+15%) would generate the greatest incremental impact (0.7%). The Markov model allows us to identify critical points of loss in the maternal care continuum. Interventions should be prioritized in rural and lower-income populations, focusing on the effective linkage between prenatal care and institutional delivery.

This study evaluated the influence of an extract rich in hederagenin obtained from Chenopodium Quinoa Willd. on cell morphology and caspase gene response in the A549 lung cancer cell line. Quinoa is notable for its high content of secondary metabolites with biological activity, with triterpene saponins being the most prominent and relevant in the field of oncology. In order to release the aglycones of interest, an initial extraction was performed with ethanol:water (1:1) followed by fractionation with n-butanol and controlled acid hydrolysis (pH 1, 80°C), purifying the final product with ethyl acetate. Quantification by high-performance liquid chromatography (HPLC) confirmed the presence of hederagenin at a concentration of 92.1 mg/L in the extract. Biological activity was determined using the MTT cell viability assay and analysis of morphological changes by hematoxylin-eosin staining. Finally, quantification by RT-qPCR revealed that there was no significant relative overexpression of caspase 3 or 7 levels, and the results of morphological analysis show that A549 cells treated with the extract generated autophagic vacuoles. These results demonstrate that the hederagenin-rich extract, under the experimental conditions evaluated, exerts its cytotoxic effect through cell death mechanisms independent of caspase activation.

Gut-brain axis disorders (GBAD) are a major public health issue in developing nations, especially Central America, where limited healthcare resources require effective disease detection and prevention. This study compares classic statistical methods and modern machine learning algorithms for identifying GBAD risk variables in Hondurans. A dataset of 1,847 12-49-year-olds was constructed using Monte Carlo simulation and epidemiological factors from regional health surveys. We used Random Forest, Gradient Boosting, Support Vector Machine, and Multilayer Perceptron Neural Network classifiers with multivariate logistic regression, chi-square analysis, and odds ratio estimates. Ten-fold stratified cross-validation calculated AUC-ROC, sensitivity, specificity, and F1-score. We found that multivariate logistic regression outperformed machine learning methods in discrimination (AUC-ROC = 0.692, 95% CI: 0.649-0.735), with Random Forest having the highest AUC (0.609). Female sex, smoking, alcohol consumption, and previous gastrointestinal diagnosis were risk factors, but physical activity was protective (OR = 0.723, 95% CI: 0.575-0.908). Traditional statistical methods may outperform advanced machine learning models in epidemiological studies with moderate sample sizes and interpretability criteria, yielding clinically useful risk estimates for primary healthcare interventions

Healthcare data fragmentation and heterogeneity pose significant challenges for reliable clinical analytics and artificial intelligence applications. This study presents a systematic data engineering pipeline designed to improve data quality, semantic normalization, and analytical readiness in clinical environments. The pipeline integrates data validation, cleansing, standardization using HL7 FHIR standards, and quality assessment modules. We evaluated the pipeline using a simulated heterogeneous clinical dataset comprising 50,000 patient records with intentionally introduced quality defects representing real-world data challenges. Quality metrics including completeness, consistency, validity, and semantic coherence were measured before and after pipeline processing. Results demonstrated substantial improvements: completeness increased from 73.2% to 98.5% (p<0.001), data consistency improved from 68.7% to 96.3% (p<0.001), duplicate records were reduced from 8.3% to 0.2%, and semantic standardization reached 97.8% conformance with FHIR resources. The pipeline successfully transformed fragmented clinical data into analysis-ready datasets suitable for advanced analytics and machine learning applications. These findings suggest that systematic data engineering approaches can significantly enhance the reliability and interoperability of clinical information systems, thereby supporting evidence-based decision-making and precision medicine initiatives.