resultados
Busca
Comunidades no DSpace
Selecione uma comunidade para navegar por suas coleções
Submissões Recentes
listelement.badge.dso-type Item , Associations between Anxiety, Depression, Chronic Pain and Oral Health-Related Quality of Life, Happiness, and Polymorphisms in Adolescents’ Genes(Universidade Federal de Minas Gerais, 2023) Ana Luiza Peres Baldiotti; Gabrielle Amaral-Freitas; Mariane Carolina Faria Barbosa; Paula Rocha Moreira; Renato Assis Machado; Ricardo Della Coletta; Michelle Nascimento Meger; Saul Martins Paiva; Rafaela Scariot; Fernanda de Morais FerreiraAdolescence is marked by changes and vulnerability to the emergence of psychological problems. This study aimed to investigate associations between anxiety/depression/chronic pain and oral health-related quality of life (OHRQoL)/happiness/polymorphisms in the COMT, HTR2A and FKBP5 genes in Brazilian adolescents. A cross-sectional study was conducted with ninety adolescents 13 to 18 years. Anxiety, depression and chronic pain were evaluated using the RDC/TMD. The Oral Health Impact Profile was used to assess oral OHRQoL. The Subjective Happiness Scale was used to assess happiness. Single-nucleotide polymorphisms in COMT (rs165656, rs174675), HTR2A (rs6313, rs4941573) and FKBP5 (rs1360780, rs3800373) were genotyped using the Taqman® method. Bivariate and multivariate logistic regression analyses were performed (p < 0.05). Chronic pain and depression were associated with feelings of happiness (p < 0.05). A significant inverse association was found between anxiety and OHRQoL (p = 0.004). The presence of minor allele C of COMT rs174675 was significantly associated with depression (p = 0.040). Brazilian adolescents with depression and chronic pain considers themselves to be less happy than others and those with anxiety are more likely to have a negative impact on OHRQoL. Moreover, the rs174675 variant allele in the COMT gene was associated with depressive symptoms in Brazilian adolescents.listelement.badge.dso-type Item , Adaptação transcultural do Parkinson's Disease Caregiver Burden Questionnaire(Universidade Federal de Minas Gerais, 2024) Ana Maria de Paula Esquárcio Louzada; Iza de Faria-Fortini; Antônio Pedro Vargas; Ana Carolina de Sousa Cruz; Paula Luciana ScalzoIntroduction. The Parkinson's Disease Caregiver Burden Questionnaire (PDCB) is a specific instrument for assessing the burden on caregivers of individuals with Parkinson's disease (PD), but its cross-cultural adaptation is necessary for its use in the Brazilian context. Objective. To perform the cross-cultural adaptation of the PDCB into Brazilian Portuguese and evaluate its content validity. Method. The study was developed in five stages: initial translation, synthesis of translations, back-translation, analysis by a committee of experts, and pre-testing with ten caregivers of people with PD followed up at a rehabilitation hospital. The content validity index (CVI) was evaluated for the criteria of clarity/comprehension, importance/relevance, and comprehensiveness. Results. The items in the adapted version of the PDCB showed semantic, idiomatic, cultural, and conceptual equivalences. The pre-test did not reveal problems regarding the interpretation of the items, demonstrating that it is a questionnaire that is easy to understand and apply. All items obtained a CVI greater than 0.9 for the three criteria evaluated. Conclusion. The PDCB-Brazil proved to be easy to understand, relevant and comprehensive, with adequate content validity and quick application for assessing the burden on caregivers of people with PD.listelement.badge.dso-type Item , Desastre ambiental na bacia do Rio Doce(Universidade Federal de Minas Gerais, 2024-01) Marcilene Miguel Cícerolistelement.badge.dso-type Item , Uma estratégia para seleção de métricas de interpretabilidade(Universidade Federal de Minas Gerais, 2026-03-20) Helton Elias de OliveiraExplaining Artificial Intelligence (AI) models is a latent contemporary need, especially because models have taken center stage in some important decisions previously restricted to human cognition. In this context, legislation such as the General Data Protection Law (LGPD) and Bill No. 2,338/2023 reinforce the need for algorithmic transparency me chanisms, accountability, and justification of automated decisions, especially in sensitive contexts such as the healthcare sector. Although the literature references a wide variety of interpretability metrics proposing contributions, this fragmentation compromises the selection of the most informative ones. This study proposes a strategy for selecting metrics focused on the quantitative evaluation of post-hoc methods, in addition to experimenting with a pragmatic sampling approach to optimize evaluations. The research began with a conceptual understanding of XAI and the limits between interpretability and explainability. This analysis allowed us to understand that interpretability is associated with quantitative analyses, based on formal and complementary metrics, while explainability extends to the qualitative field, involving human validation of explanations, contextual understanding, and adequate communication of results. We adopted two ensemble-based classifiers and evaluated their predictive performance in four public datasets, in addition to a dataset of medical examinations of patients with suspected sepsis from MIMIC. While Random Forest uses the model combination strategy by bagging, XGBoost is based on boosting. Both had hyperparameters adjusted using Optuna, employing TPE. The strategy systematizes evaluating models through five widely used post-hoc methods; These are: Anchor (Anchor-based Explanations), LIME (Local Interpretable Model-agnostic Explanations), Permutation Feature Importance, SHAP (SHapley Additive exPlanations), and Surrogate Decision Tree. The choice of these methods prioritized diversifying the approaches, including: perturbations around instances, attribute importance, rule extraction, additive assignment, and surrogate models, allowing for a comprehensive comparative analysis from different explanatory perspectives. Considering the selected methods, we coded twelve complementary quantitative metrics in Python, which allowed us to evaluate them individually from different perspectives. The selection of calculated metrics was based on diversity, in order to address different aspects of the relationship between the post-hoc methods and the evaluated models, prioritizing metrics with consistent references in the literature. This preliminary analysis resulted in the selection of the metrics fidelity, infidelity, faithfulness, completeness, selectivity, simplicity, consistency, sufficiency, stability, robustness, soundness, and directional soundness. The central contribution of this research was the identification of informational overlap between the calculated metrics, with analysis conducted using the Friedman test, followed by Spearman correlation analyses for metrics that showed similar behavior. Complementarily, the Nemenyi test contributed in scenarios where the metrics, in a homogeneous way, differed statistically. We provide a Scorecard that offers a conceptual model for metric selection when analyzing grouped or individual data. The reduction in computational cost is based on a sampling methodology via prototypical instance selection. This approach relies on clustering and prototype selection methods, such as ProtoDash and K-center. The goal is to ensure hat the sample represents the overall behavior of the data, mitigating computational and cognitive costs without loss of explanatory fidelity. The quantitative evaluation of metrics, as well as the simplification of the sample space, allowed for the analysis of the coherence, utility, and reliability of explanations in the sensitive and critical domain of health. The results indicate that the structured integration between quantitative metrics and prototypical analysis constitutes a consistent approach for evaluating explanations in XAI.listelement.badge.dso-type Item , Integrated observation-model analysis to quantify and improve fugitive PM10 emission estimates from open-pit iron ore mining(Universidade Federal de Minas Gerais, 2026-03-20) Otavio Medeiros SobrinhoMinas Gerais é responsável por mais de 70% da produção brasileira de minério de ferro, concentrada majoritariamente no Quadrilátero Ferrífero, onde minas a céu aberto de grande porte impactam severamente a qualidade do ar local através de emissões fugitivas de material particulado (MP10). Quantificar a contribuição dessas fontes difusas é complexo devido à baixa densidade de monitoramento e à representação inadequada de processos mineiros (como tráfego em estradas não pavimentadas, detonação e manuseio de pilhas) em inventários globais, o que resulta em subestimativas sistemáticas em modelos de transporte químico. Esta tese integra modelagem meteorológica e química (WRF-SMOKE-CMAQ), ferramentas estatísticas direcionais e modelagem inversa (AERMOD) para diagnosticar, quantificar e refinar as estimativas de emissões de MP10 da mineração. A pesquisa estruturou-se em três estudos. No primeiro, focado na Região Metropolitana de Belo Horizonte (MABH), o uso de gráficos polares anulares, bivariados e função de probabilidade condicional (CPF) identificou assinaturas distintas de fontes veiculares, industriais e de mineração. Evidenciou-se que a queima de biomassa, especialmente em setembro de 2021, atua como um amplificador episódico crítico, embora as fontes locais mantenham os níveis de fundo elevado. No segundo estudo, em Itabira, simulações de alta resolução (1 km) revelaram que o modelo CMAQ subestima o MP10 em aproximadamente -83% (NMB médio) ao utilizar o inventário global EDGAR-HTAP v2. Uma filtragem meteorológica (ΔWind Dir. < 25° and ΔWind Speed < 0,5 m s-1) provou que o viés negativo persiste mesmo sob condições de vento bem simuladas, confirmando que a falha reside na incompletude estrutural do inventário e não em erros de fase meteorológica. Testes de sensibilidade com poeira eólica (wind-blown dust) mostraram melhorias marginais de apenas 3% no viés, reforçando que os processos antropogênicos ativos são as principais fontes omitidas. No terceiro estudo, aplicou-se modelagem inversa com o modelo AERMOD para derivar fatores de correção. Identificou-se que as emissões de mineração precisam ser multiplicadas por fatores entre 42 e 48 para alinhar as simulações às observações de estações próximas à mina, resultando em um fluxo superficial ajustado de ~1,4 × 10-9 kg m-2 s-1. A propagação desse ajuste para o modelo regional CMAQ (cenário MIN-ADJ) reduziu o viés médio da rede de -60% para -34%. Conclui-se que a integração de monitoramento de proximidade com modelagem inversa oferece um caminho prático para aprimorar inventários e subsidiar o licenciamento ambiental. O trabalho destaca que a ausência de dados precisos caracteriza uma injustiça ambiental, pois compromete a proteção de comunidades mineradoras vulneráveis frente aos novos padrões da Resolução CONAMA 506/2024 e da Lei 14.850/2024.