Detecção automática de complexos-K no sinal EEG humano baseada em análise espectral multitaper.

• Ano de defesa: 2019
• Autor(a) principal: OLIVEIRA, Gustavo Henrique Batista Santos
• Orientador(a): COUTINHO, Luciano Reis
• Banca de defesa: COUTINHO, Luciano Reis , BRAZ JÚNIOR, Geraldo , TEIXEIRA, Silmar Silva
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM CIÊNCIA DA COMPUTAÇÃO/CCET
• Departamento: DEPARTAMENTO DE INFORMÁTICA/CCET
• País: Brasil
• Palavras-chave em Português: Eletroencefalograma; complexo-K; espectrograma multitaper; estudo do sono; análise espectral
• Palavras-chave em Inglês: Electroencephalogram; k-complex; multitaper spectrogram; sleep study; spectral analysis
• Área do conhecimento CNPq: Engenharia de Software.
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/2738

In this work it is proposed a novel method for automatic k-complex (KC) detection in human sleep EEG, named MT-KCD. Like most methods, MT-KCD codifies some rules used by human experts such as KC characterization as 0-4 Hz waveform standing out from the background, with peak-to-peak amplitude ≥ 75 𝜇𝑉 and duration ≤ 2 seconds. The MT-KCD novelty when compared to existing methods is the usage of multitaper spectral analysis to characterize the KC as 0-4 Hz waveform standing out from the background. The EEG multitaper spectral analysis is a recntly proposed tecnique by researches as a complement to the traditional hypnogram to sleep staging. The MT-KCD consists in three phases: pre-processing, candidates extraction and candidates elimination. In pre-processing phase, EEG multitaper spectrogram is computed. In sequence, the multitaper spectrogram is used to identify regions where possible have KCs occurrences. Lastly, candidates waveform which satisfy duration and peak-to-peak amplitude criterias are marked as KC. MT-KCD was evaluated using a public KC database known as DREAMS. Results have shown that MT-KCD improves detection metrics, especially F1 and F2 scores (harmonic averages of recall and precision), when compared to existing methods. F1 and F2 scores of MT-KCD on DREAMS were greater than 75%, in most of the evaluation scenarios, outperforming other methods. In regards to recall and precision, MT-KCD is comparable to existing methods in recall and precision, but presenting a more balanced relation between these metrics (F1 and F2 scores).