Detecção de transtorno mental via EEG, microestados e redes neurais de grafos

• Ano de defesa: 2025
• Autor(a) principal: Candeia, Daniel Ribeiro
• Orientador(a): Ciarelli, Patrick Marques
• Banca de defesa: Côco, Klaus Fabian , Tello, Richard Junior Manuel Godinez
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; Mestrado em Engenharia Elétrica
• Programa de Pós-Graduação: Programa de Pós-Graduação em Engenharia Elétrica
• Departamento: Centro Tecnológico
• País: BR
• Palavras-chave em Português: Rede Neural de Grafos; Microestados de EEG; Grafos; Eletroencefalograma; Transtornos Mentais; Graph Neural Network; EEG Microstates; Graphs; Electroencephalogram; Mental Disorders
• Área do conhecimento CNPq: Engenharia Elétrica
• Link de acesso: http://repositorio.ufes.br/handle/10/20640

Electroencephalogram (EEG) is a non-invasive and cost-effective technique widely used to study brain activity and diagnose neurological disorders. However, visual analysis of EEG signals is complex and requires expertise, highlighting the need for automated diagnostic support systems. In this context, this study proposes a graph-based neural network model for detecting mental disorders using EEG signals, leveraging microstate analysis. The proposed model integrates graph neural networks (GNNs) with microstate analysis, which captures transient and stable patterns of brain activity. The TUH Abnormal EEG Corpus (TUAB) dataset, containing normal and abnormal EEG signals, was used. The process included the extraction of microstates, the construction of graphs based on Spearman correlation between EEG channels, the extraction of features from EEG signals, and the application of Principal Component Analysis (PCA) to reduce the dimensionality of these features. Three GNNs were trained, each associated with signals from each microstate, and their outputs were combined using an ensemble technique. The final model achieved an accuracy of 97.46% on the test set, outperforming existing results of methods in the literature. The results highlight the effectiveness of the proposed approach, demonstrating the potential of GNNs and microstate analysis for detecting mental disorders from EEG signals