Classificação de estágios de sono através da aplicação de transformada wavelet discreta sobre um único canal de eletroencefalograma

• Autor(a) principal: Silveira, Thiago Lopes Trugillo da
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Manancial - Repositório Digital da UFSM
• dARK ID: ark:/26339/001300000jknp
• Texto Completo: http://repositorio.ufsm.br/handle/1/15433
• topic: Transformada wavelet discreta (DWT); Classificação do sono; Sinais de eletroencefalograma (EEG); Discrete wavelet transform (DWT); Sleep classification; Electroencephalogram (EEG) signals; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

The correct sleep stage classification allows sleep experts to diagnose and treat disorders such as apnea, narcolepsy and insomnia. Such task is classically performed by sleep medicine experts, where one or more physiological signals are visually inspected. Since electroencephalogram (EEG) signals are considered good indicators for sleep analysis, they are widely used for sleep stage scoring. Although, the adequate sleep classification for a single night of sleep can demand from two to four hours of analysis, being frequently performed by two experts. The current study presents a novel decision support system, aiming to facilitate this experts’ task. The proposed methodology is based on the multi-resolution analysis of a single EEG channel through the application of the discrete wavelet transform (DWT). Methodologies which consider only one EEG channel for sleep scoring have reduced computational cost and the related acquisition equipments are easier to use in comparison with the multi-channel ones. Afterwards the signal decomposition by DWT, statistical features from sleep related brain rhythms are extracted and feed a classifier. Random forests are used for classification task in the current study. A set of 39 signals corresponding to 20 volunteers of a public database is considered. The performance of the proposed method is evaluated through techniques such as cross-validation, where accuracies keep higher than 90% and kappa coefficient higher than 0.8 are achieved for 2 to 6 states of sleep stages. The introduced method achieves better, or at least comparable, results when compared with state-of-the-art studies in all analyzed scenarios.