Métodos de extração de atributos para detecção de crises epiléticas: uma abordagem comparativa

• Ano de defesa: 2019
• Autor(a) principal: Silva Júnior, Júlio Peixoto da
• Orientador(a): Cavalcante, Charles Casimiro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Teleinformática; Diagnóstico por computador; Epilepsia - Classificação; Epilepsy; Features extraction; Power spectral density; Linear predictive coding; Extreme learning machine; Minimal learning machine; Random kitchen sinks; Suport vector machine
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/40928

Epilepsy is a brain disorder characterized by recurrent epileptic seizures that affects approximately 50 million people worldwide, making it one of the most common neurological diseases. The electroencephalogram (EEG) is an electrophysiological monitoring method that records the electrical activity of the brain and is widely used in the detection and analysis of epileptic seizures. However, it is often difficult to identify subtle but critical changes in the EEG waveform through visual inspection.The EEG signals are nonlinear and non-stationary in nature, which contribute to further complexities related to their manual interpretation and detection of normal and abnormal (interictal and ictal) activities. Therefore, there is a need for a computer aided diagnostic system to automatically identify the anormal activities. It was found that the use of the attribute vector with the MFCC coefficients presented the best performance among the extraction methods, obtaining a sensitivity of 97%, specificity of 98% and a mean of 99.8% in all classifiers.In this dissertation we propose a system to aid the patient diagnosis in specific, in which a comparison was made between four methods of extraction of aracteriststicas: Power Spectral Density, Linear Predictive Coding, Mel-Frequency Cepstral Coefficients and covariance matrix. These extraction methods were combined in scenarios with three types of randomized pattern classifiers: Extreme Learning Machine, Random Kitchen Sinks and Minimal Learning Machine. And for the purpose of comparison was used the classifier Support Vector Machine. In the simulations performed, the proposed scenarios used files with about one hour (in some cases up to four hours), were used and the results pointed out that the random classifiers are dependent on the method of extraction of characteristics used. It was found that the use of the attribute vector with MFCC coefficients presented the best performance among the extraction methods, obtaining a sensitivity of 97%, specificity of 98% and a mean of 99.8% in all classifiers.