Emprego da análise de componentes de grupo relacionadas à tarefa na decomposição de potenciais relacionados a eventos
Arquivos
Data
2023-08-16
Autores
Couto, Bruno Andry Nascimento 
Orientadores
Casali, Adenauer Girardi
Tipo
Dissertação de mestrado
Título da Revista
ISSN da Revista
Título de Volume
Resumo
Este estudo tem como objetivo explorar a eficácia do método de Análise de Componentes de Grupo Relacionadas à Tarefa (gTRCA) na extração de componentes reprodutíveis a partir de dados de Potenciais Relacionados a Eventos (ERPs) e na interpretação da natureza complexa das formas de onda dos ERPs. O estudo é conduzido em três fases, cada uma focando em um aspecto diferente da aplicação do método. A primeira fase consiste na análise e validação da gTRCA usando conjuntos de dados simulados, revelando detalhes sobre o funcionamento do método e sua robustez em vários cenários. Nesta etapa, um novo teste estatístico, implementado a partir do deslocamento circular dos sinais de cada indivíduo, também foi introduzido para medir a reprodutibilidade a nível de grupo das componentes, provando-se útil na análise dos ERPs. A segunda fase envolve a aplicação da gTRCA a conjuntos de dados reais de ERPs, obtidos de 40 indivíduos submetidos a um protocolo de Mismatch Negativity (MMN) auditivo passivo do tipo oddball. O objetivo era investigar o potencial do método na implementação de classificadores para tipos de potenciais distintos: padrão e desviante. A terceira fase estende a aplicação a outro conjunto de dados reais de 16 sujeitos submetidos a um protocolo de Estimulação Magnética Transcraniana (TMS) no córtex motor esquerdo, visando caracterizar os Potenciais Evocados por TMS (TEPs). Como resultado, na primeira fase mostramos - em dados simulados - que o método é capaz de identificar as componentes que são reprodutíveis intra e entre sujeitos desde que seja combinado a técnicas estatísticas apropriadas. Na segunda fase, encontramos que a gTRCA foi capaz de extrair componentes dos potenciais auditivos MMN de forma a discriminar os diferentes tipos de estímulo com alta acurácia (mediana de 90%). Finalmente, mostramos que a técnica identifica múltiplas componentes reprodutíveis em nível de grupo em TEPs motores e que possuem características nos domínios do espaço, tempo e frequência coerentes com a literatura descrevendo tais potenciais. Deste modo, esta investigação multifacetada não só aprofunda nossa compreensão dos ERPs como ferramentas de investigação em neurociências, mas também destaca o potencial de técnicas de aprendizado de máquina como a gTRCA na análise de ERPs. Os resultados deste trabalho ilustram a robustez e aplicabilidade da gTRCA, posicionando-a como uma ferramenta valiosa na análise e interpretação de ERPs.
This study aims to explore the efficacy of the group Task-Related Component Analysis (gTRCA) in extracting reproducible components from Event-Related Potential (ERP) and in interpreting the complex nature of ERP waveforms. The study is conducted in three phases, each focusing on a different aspect of the method's application. The first phase consists of an analysis and validation of gTRCA using simulated datasets, revealing detailed insights into the workings of the method and its robustness in various scenarios. In this phase, a novel statistical test, based on the circular-shifting of each individual evoked potential, was also introduced to measure component consistency, proving useful in ERP analysis. The second phase involves applying gTRCA to real ERP datasets, obtained from 40 individuals subjected to an auditory mismatch negativity protocol. The goal was to investigate the method's potential in implementing classifiers for distinct potential types: standard and deviant. The third phase extends the application to another real dataset from 16 subjects undergoing Transcranial Magnetic Stimulation (TMS) in the left motor cortex, aiming to characterize TMS-Evoked Potentials (TEPs). As a result, in the first phase, we demonstrated - using simulated data - that the method can identify components that are reproducible both within and between subjects, provided it is combined with appropriate statistical techniques. In the second phase, we found that gTRCA was able to extract components from the MMN auditory potentials in a way that could differentiate between different types of stimuli with high accuracy (a median of 90%). Finally, we showed that the technique identifies multiple reproducible components at a group level in motor EPs, and these have characteristics in the domains of space, time, and frequency that are consistent with the literature describing such potentials. Therefore, this multifaceted investigation not only deepens our understanding of ERPs as research tools in neuroscience, but also highlights the potential of machine learning techniques like gTRCA in the analysis of ERPs. Results from these phases illustrate the robustness and utility of gTRCA, positioning it as a valuable tool in the analysis and interpretation of ERPs.
This study aims to explore the efficacy of the group Task-Related Component Analysis (gTRCA) in extracting reproducible components from Event-Related Potential (ERP) and in interpreting the complex nature of ERP waveforms. The study is conducted in three phases, each focusing on a different aspect of the method's application. The first phase consists of an analysis and validation of gTRCA using simulated datasets, revealing detailed insights into the workings of the method and its robustness in various scenarios. In this phase, a novel statistical test, based on the circular-shifting of each individual evoked potential, was also introduced to measure component consistency, proving useful in ERP analysis. The second phase involves applying gTRCA to real ERP datasets, obtained from 40 individuals subjected to an auditory mismatch negativity protocol. The goal was to investigate the method's potential in implementing classifiers for distinct potential types: standard and deviant. The third phase extends the application to another real dataset from 16 subjects undergoing Transcranial Magnetic Stimulation (TMS) in the left motor cortex, aiming to characterize TMS-Evoked Potentials (TEPs). As a result, in the first phase, we demonstrated - using simulated data - that the method can identify components that are reproducible both within and between subjects, provided it is combined with appropriate statistical techniques. In the second phase, we found that gTRCA was able to extract components from the MMN auditory potentials in a way that could differentiate between different types of stimuli with high accuracy (a median of 90%). Finally, we showed that the technique identifies multiple reproducible components at a group level in motor EPs, and these have characteristics in the domains of space, time, and frequency that are consistent with the literature describing such potentials. Therefore, this multifaceted investigation not only deepens our understanding of ERPs as research tools in neuroscience, but also highlights the potential of machine learning techniques like gTRCA in the analysis of ERPs. Results from these phases illustrate the robustness and utility of gTRCA, positioning it as a valuable tool in the analysis and interpretation of ERPs.
Descrição
Citação
Couto, B.A.N. & Casali, A.G. (2023). Emprego da Análise de Componentes de Grupo Relacionadas à Tarefa na Decomposição de Potenciais Relacionados a Eventos.