Geometric deep learning for functional neuroimaging analysis

• Ano de defesa: 2021
• Autor(a) principal: Marcon, Matheus Zampieri
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Escola Politécnica; Brasil; PUCRS; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Artificial Intelligence; Neuroimaging; fMRI; Deep Neural Networks; Inteligência Artificial; Neuroimagem; Redes Neurais Profundas; CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/9762

The study of the human brain connectome, a complex set of cerebral network relationships associating structure and functionality, has seen a growing interest in the field of neuroimaging over the last decade. Deep learning techniques constitute the state-ofthe-art for neuroimaging classification tasks on different neurological disorders, providing in-depth analysis into the inherent characteristics of brain activation and connectivity without the need for prior feature selection. However, convolutional operations of traditional deep networks affect fixed regions of elements during learning, whereas connectome data is best represented in the form of graphs, with spatially dispersed elements. We make use of geometric deep learning (GDL) for the analysis of whole-brain functional magnetic resonance imaging (fMRI) connectome data to identify and extract high-level feature representations of the cerebral network dynamics involved in human cognition. Our findings suggest that GDL techniques can outperform state-of-the-art models for classification of fMRI data while providing a simple framework for result analysis.