Rastreamento não invasivo para diabetes tipo 2

• Ano de defesa: 2015
• Autor(a) principal: Ribeiro, Áurea Celeste da Costa
• Orientador(a): BARROS FILHO, Allan Kardec Duailibe
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
• Departamento: DEPARTAMENTO DE ENGENHARIA DA ELETRICIDADE/CCET
• País: Brasil
• Palavras-chave em Português: Processamento de sinais; Rastreamento; Diabetes; Não Invasivo; Signal processing; Screening; Non invasive
• Área do conhecimento CNPq: Bioengenharia
• Link de acesso: http://tedebc.ufma.br:8080/jspui/handle/tede/1607

The type 2 diabetes screening has become an important resource due to the increase in this disease in the modern world, it is estimated that there are 385 millions of diabetics in worldwide and that 46% of this number are unaware of their condition. This complicates their treatment and many patients at diagnosis already present any complications due to lack this in the early stages of diabetes. Researchers have discussed the e ectiveness of type 2 diabetes screening, for example: The Brazil made a screening in 2001 and it was considered an unnecessary cost of almost 40 million. The tracking of type 2 diabetes has become an important resource due to the large increase in this disease in the modern world, it is estimated that there are 385 million diabetics worldwide and that 46% of this number are unaware of their condition. This complicates their treatment and many patients the diagnosis already present any complications due to lack this in the early stages of diabetes. There were discussions about the effectiveness of screening for type 2 diabetes, in Brazil for example, the last scan was considered unnecessary cost of almost 40 million. Simplest and most effective methods of screening are studied, such as the US and China that use some non-invasive methods to calculate the risk of diabetes. This study proposes a non-invasive screening method based on eficient coding technique to extract features of a Brazilian database (HIPERDIA) to form a new concise representation thereof, with the decrease of redundancy. The main hypothesis worked at this stage was the pursuit of independent components, which possibly it were present at the formation of the disease. Thus, the original data were decomposed by the independent component analysis method. In the classification stage to ensure discrimination between classes was used the method of support vector machines for one class. Tests were done to check the performance of the classifier after the feature extraction phase, and showed that it increases the performance of support vector machine to one class in making the discrimination between diabetics and non-diabetics. Results were reached (100%) with the combination of certain characteristics, and the method shows promise in obtaining a non invasive type 2 diabetes screening. Other tests were done to determine the influence of each non invasive marker in the final result and the generality of the method using other databases, as t of the Pima Indians and African Americans data sets. Then, reducing the number of features used to train the method and testing whether all possible combinations among the remaining, removing one by one, a total of 12,910 possibilities. It was observed the characteristics or markers that most affected the final outcome were age and characteristics related to body fat. Testing the generality of the method in other databases found that the method works best with balanced data set.