A carregar...
Publicação
Análise estatística espacial e previsão clínica de aneurismas da aorta torácica ascendente
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 7.46 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
A deteção precoce do risco de rutura de aneurismas da aorta permite intervir cirurgicamente em tempo útil. O tamanho do aneurisma é avaliado pelo diâmetro; segundo algumas diretrizes clínicas, recomenda-se a reparação do aneurisma para diâmetros superiores a 50 mm em mulheres e superiores a 55 mm em homens. No entanto, tem-se evidenciado que este parâmetro não é suficiente como indicador de cirurgia. Neste trabalho procura-se contribuir para o estudo deste problema; para tal, realiza-se uma análise exploratória com bases de dados de doentes com aneurismas localizados na aorta ascendente. Aplicam-se técnicas de estatística espacial e de Machine Learning para a caraterização.
Relativamente às técnicas de estatística espacial, utilizou-se o cálculo de variogramas para o estudo espacial dos diâmetros máximos na aorta ascendente e a técnica de krigagem ordinária para a estimação de novos diâmetros máximos. No âmbito das técnicas de Machine Learning, recorreu-se, no caso da aprendizagem supervisionada, à regressão logística e, na aprendizagem não supervisionada, ao método k-means, com o objetivo de estudar se, a partir das variáveis clínicas, seria possível determinar se os doentes necessitam ou não de cirurgia.
Verificou-se que o variograma sugere a localização do aneurisma na aorta e que, a partir dos seus parâmetros, é possível estudar propriedades biomecânicas da aorta, como a
elasticidade e a rigidez local. Por sua vez, a técnica de krigagem ordinária revelou-se útil na estimação de diâmetros máximos em instantes que não foram objeto de amostragem.
Além disso, a análise das variáveis clínicas permitiu concluir que o diâmetro máximo é a variável mais relevante na decisão de operar, em consonância com a literatura e com as recomendações atualmente vigentes.
The early detection of the risk of aortic aneurysm rupture allows for timely surgical intervention. The size of the aneurysm is assessed by its diameter; according to certain clinical guidelines, surgical repair is recommended for diameters greater than 50 mm in women and 55 mm in men. However, evidence has shown that this parameter alone is not sufficient as an indicator for surgery. This study aims to contribute to the understanding of this problem by conducting an exploratory analysis using databases of patients with aneurysms located in the ascending aorta. Techniques from spatial statistics and Machine Learning are applied for characterisation. Regarding spatial statistics techniques, variogram computation was employed for the spatial analysis of maximum diameters in the ascending aorta, and ordinary kriging was used to estimate new maximum diameters. Within the scope of Machine Learning techniques, logistic regression was applied as the supervised learning method, and k-means as the unsupervised method, with the aim of studying whether, based on clinical variables, it would be possible to determine whether patients require surgery. The variogram was found to suggest the location of the aneurysm within the aorta; moreover, its parameters make it possible to study biomechanical properties of the aorta, such as elasticity and local stiffness. In turn, ordinary kriging proved useful for estimating maximum diameters at time points that were not sampled. Furthermore, the analysis of clinical variables led to the conclusion that maximum diameter is the most relevant variable in the decision to operate, in agreement with the literature and with current clinical guidelines.
The early detection of the risk of aortic aneurysm rupture allows for timely surgical intervention. The size of the aneurysm is assessed by its diameter; according to certain clinical guidelines, surgical repair is recommended for diameters greater than 50 mm in women and 55 mm in men. However, evidence has shown that this parameter alone is not sufficient as an indicator for surgery. This study aims to contribute to the understanding of this problem by conducting an exploratory analysis using databases of patients with aneurysms located in the ascending aorta. Techniques from spatial statistics and Machine Learning are applied for characterisation. Regarding spatial statistics techniques, variogram computation was employed for the spatial analysis of maximum diameters in the ascending aorta, and ordinary kriging was used to estimate new maximum diameters. Within the scope of Machine Learning techniques, logistic regression was applied as the supervised learning method, and k-means as the unsupervised method, with the aim of studying whether, based on clinical variables, it would be possible to determine whether patients require surgery. The variogram was found to suggest the location of the aneurysm within the aorta; moreover, its parameters make it possible to study biomechanical properties of the aorta, such as elasticity and local stiffness. In turn, ordinary kriging proved useful for estimating maximum diameters at time points that were not sampled. Furthermore, the analysis of clinical variables led to the conclusion that maximum diameter is the most relevant variable in the decision to operate, in agreement with the literature and with current clinical guidelines.
Descrição
Mestrado em Estatística, Matemática e Computação, apresentda à Universidade Aberta
Palavras-chave
Machine Learning Estatística espacial Aneurismas da aorta torácica Krigagem ordinária
Contexto Educativo
Citação
Editora
Licença CC
Sem licença CC