Uma nova abordagem baseada em autovalores para a estimação de posições de fontes cerebrais utilizando sinais eletroencefalográficos

• Autor(a) principal: Cruz, Lucas Fiorini
• Data de Publicação: 2018
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFG
• dARK ID: ark:/38995/0013000004vqx
• Texto Completo: http://repositorio.bc.ufg.br/tede/handle/tede/8793
• topic: EEG; DOA EEG; Localização de fontes cerebrais; Localização de focos epileptogênicos; EEG; DOA EEG; Brain sources localization; Epileptogenic foci localization; BIOENGENHARIA::PROCESSAMENTO DE SINAIS BIOLOGICOS

Electroencephalography (EEG) measurements are widely used in clinical assessments for research due to its noninvasive nature and for providing several informations on the neural activity associated to both neural functions and disorders, including epilepsy syndromes. In cases related with partial epilepsy, surgical interventions are recommended and the accurate location of the seizure becomes a sine qua non prerequisite for those procedures. Brain source position estimation can help in the selection and classification of brain spots. In this sense, for control purposes, this work begins with the development of a mathematical model that is more electromagnetically representative than the usual model, presenting different aggregate characteristics such as refraction and, especially, frequency dependent attenuation of the wave. We also propose a new method that estimates the source positions from spectral peaks produced by the eigenvalues of the sum of the spatial covariance matrix of the EEG signals and a spatial covariance matrix related to a simulated source that is numerically swept throughout every point on different horizontal layers of the brain. The key approach was to select the eigenvalues that were less affected by the noise and use them to produce the search spectrum. In order to assess the accuracy and robustness of the proposed method, we compared its RMSE (Root Mean Square Error) performance at different SNRs (Signal-to-Noise Ratio) to those of MUSIC (Multiple Signal Classification), a method based on orthogonal subspaces, and NSF (Noise Subspace Fitting), a method based on subspace fitting. The results were produced for both the usual and proposed signal model in order to evaluate their accuracy. Subsequently, the signal models were compared after spatial filtering, aiming the determination of the waveform of a particular source. The proposed approach presents the lowest threshold SNR and the highest accuracy under noisy conditions for all analyzed cases and for both models. The new approach for the signal model made the estimation more accurate in all the studied cases, besides providing greater accuracy on spatial filtering, when compared to the usual model.