Otimização multiobjetivo aplicada na identificação de parâmetros para análise eletroencefalográfica
Ano de defesa: | 2019 |
---|---|
Autor(a) principal: | |
Orientador(a): | |
Banca de defesa: | |
Tipo de documento: | Tese |
Tipo de acesso: | Acesso aberto |
Idioma: | por |
Instituição de defesa: |
Universidade Federal de Uberlândia
Brasil Programa de Pós-graduação em Engenharia Mecânica |
Programa de Pós-Graduação: |
Não Informado pela instituição
|
Departamento: |
Não Informado pela instituição
|
País: |
Não Informado pela instituição
|
Palavras-chave em Português: | |
Link de acesso: | https://repositorio.ufu.br/handle/123456789/24923 http://dx.doi.org/10.14393/ufu.te.2019.324 |
Resumo: | In medicine experts currently spends a long time in the analysis of tests that do not present any anomalies. For the case of the neurologist, more than 80% of the exams are good. Another problem is that the diagnostic error is very common in the neurology. Due to the time spent in analysis of normal EEG, the high percentage of errors and the consequence that to false positive diagnosis can affect to human health and life quality, all research aimed at minimizing errors in EEG's diagnosis should be welcomed, especially those involving Artificial Intelligence. In this work, to aid the EEG expert, is presented a methodology based on artificial intelligence capable of separating EEG signals into signs with and without anomalies with a high probability of success. For this was used a public available database of Bern-Barcelona University consisting of signs with ictal presence and without the presence of ictal. For the evaluation of the developed procedures was used the metrics: sensitivity, specificity, accuracy, predicted positive values, predicted negative values and Mathew correlation coefficient. Two classification procedures (QDA and SVM) and four multi-objective optimization algorithms (MOPSO, PESA2, SPEA2 and NSGA-II) were used for discriminant analysis of 1180 symptom-seeking parameters to classify EEG signals into focal and non-focal ones. Signal analysis techniques, such as classical statistical analysis (RMS value and kurtosis, for example), Wavelet and Hilbert-Huang transforms, envelope analysis and entropy and k-NEO calculations were used to choice the best parameters to classification purpose. The use of the NSGA-II and objective functions with penalization, proposed in this work, resulted in sensitivity values of 94%, specificity of 82% and an accuracy of 87% for the validation and test databases. |