Título:	Abordagens univariáveis e multivariáveis de índices autonômicos cardíacos em modelos preditivos de diabetes tipo 2 : análises com técnicas estatísticas e de aprendizado de máquina
Autor(es):	Bernardes, Alex Alves
Orientador(es):	Oliveira, Flavia Maria Guerra de Sousa Aranha
Assunto:	Processamento de sinais biomédicos Diabetes mellitus tipo 2 Aprendizado de máquina Análise estatística
Data de publicação:	18-Mar-2025
Data de defesa:	16-Ago-2024
Referência:	BERNARDES, Alex Alves. Abordagens univariáveis e multivariáveis de índices autonômicos cardíacos em modelos preditivos de diabetes tipo 2: análises com técnicas estatísticas e de aprendizado de máquina. 2024. 109 f. Dissertação (Mestrado em Engenharia Biomédica) — Universidade de Brasília, Brasília, 2024.
Abstract:	Type 2 Diabetes Mellitus is a metabolic disease whose incidence occurs mainly in adults and is continuously growing worldwide. Its symptoms affect the autonomic nervous system (ANS), causing autonomic dysfunctions. Several pieces of evidence indicate that it is possible to predict the presence of the disease through quantitative assessments of autonomic activity on cardiac regulation. This work aims to determine quantitative indices of the ANS, using univariate and multivariate techniques, that can differentiate diabetes mellitus individuals (DMI) and control group (CG) using physiological signals obtained non-invasively such as electrocardiogram (ECG), continuous blood pressure (BP), and respiratory flow (RF). Additionally, using an insightful approach, the goal is to evaluate the effectiveness of different combinations of autonomic indices for use as attributes in machine learning algorithms for predicting DMI and CG. Initially, the study investigates the use of predictive models employing autonomic indices obtained from a single signal in isolation, such as ECG or BP. Subsequently, predictive models are developed using multivariable indices, which require the combination of physiological signals complementary to the ECG, relating them in parametric models where the causality constraint between the signals is imposed. The data for this study were obtained from the public database “Cerebromicrovascular Disease in Elderly with Diabetes”. The dataset consists of ECG, continuous BP, and RF data from individuals aged 55 to 75 years, with and without type 2 diabetes. The groupings of indices were based on the physiological signals of origin and mathematical extraction methodologies. Special groupings were proposed using statistical techniques such as correlation analysis and variance analysis, in addition to machine learning techniques. Four classification algorithms were used: 1) support vector machines (SVM), 2) decision tree (DT), 3) k-nearest neighbor (KNN), and 4) logistic regression (LR). Eight sets were obtained from the groupings of autonomic indices: A) VFC/HRV - indices exclusively related to heart rate variability (HRV); B) VPA/BPV - spectral indices of blood pressure variability (BPV); C) FRF - combination of FRF indices obtained from the spectral analysis of HRV, BPV, and system identification indices by impulse response (IR); D) RI/IR - system identification indices by IR; E) VFC/HRV + VPA/BPV - the combination of indices in sets A and B; F) VFC/HRV + RI/IR - the combination of indices in sets A and D; G) Special 1 – indices selected based on variance analysis results of the indices; H) Special 2 - indices that showed high performance in prediction with individual indices. The metrics from the machine learning models using autonomic indices indicated that the grouping of indices associated with HRV in combination with other groupings of indices tends to increase the accuracy, precision, and sensitivity of the models. However, the grouping of indices associated exclusively with HRV performed worse than the other groupings in predicting diabetic individuals. The groupings of indices called special, G and H, stood out by obtaining the best accuracy and F1-score results among the other groupings in 3 of the AAM used. The SVM algorithm presented the following metrics for grouping G: accuracy (A), precision (P), sensitivity (S), and F1-score (F1): A=80.72%; P=79.47%; S=97.14%; F1=87.26%. While grouping H presented the following results: A=78.78%; P=77.53%; S=94.28%; F1=84.78%. For the KNN algorithm, grouping G presented: A=80.72%; P=80.55%; S=93.80%; F1=86.5%. Grouping H: A=80.60%; P=79.47%; S=94.28%; F1=85.94%. For the LR algorithm, grouping G: A=72.90%; S=88.57%; P=76.36%; F1=81.28%. Grouping H: A=80.60%; P=80.47%; S=91.42%; F1=85.47%. The performance divergence occurred with the DT algorithm, where the performance metrics were led by grouping E, containing univariate indices of HRV and BPV. The metrics for grouping E are: A=78,21%; P=86,66%; S=84,0%; F1=83,32%. Among the individual indices evaluated, multivariate indices associated with the dynamic gain of cardiorespiratory coupling (CRC) presented more robust results, such as the dynamic gain of high-frequency 0.15 - 0.4Hz (HF) from CRC combined with baroreflex mechanism coupling, with A=80.75% and F1=84.89%. The same trend appears in the groupings of indices, where the most indices of groupings G and H are related to the CRC, in combination with univariable indices exclusively associated with HRV. The special groupings, using feature selection, showed a significant improvement in the prediction of DMI and CG compared to groupings of indices purely grouped by signal origin and extraction methodology. The special groupings also showed an improvement in some metrics for predicting individuals compared to using individual indices, but the improvement was less significant.
Unidade Acadêmica:	Faculdade de Ciências e Tecnologias em Engenharia (FCTE) – Campus UnB Gama
Informações adicionais:	Dissertação (mestrado)—Universidade de Brasília, Faculdade UnB Gama, Programa de Pós-Graduação em Engenharia Biomédica, 2024..
Programa de pós-graduação:	Programa de Pós-Graduação em Engenharia Biomédica
Licença:	A concessão da licença deste item refere-se ao termo de autorização impresso assinado pelo autor com as seguintes condições: Na qualidade de titular dos direitos de autor da publicação, autorizo a Universidade de Brasília e o IBICT a disponibilizar por meio dos sites www.unb.br, www.ibict.br, www.ndltd.org sem ressarcimento dos direitos autorais, de acordo com a Lei nº 9610/98, o texto integral da obra supracitada, conforme permissões assinaladas, para fins de leitura, impressão e/ou download, a título de divulgação da produção científica brasileira, a partir desta data.
Aparece nas coleções:	Teses, dissertações e produtos pós-doutorado

Os itens no repositório estão protegidos por copyright, com todos os direitos reservados, salvo quando é indicado o contrário.