Brain connectivity-patterns representation based on electroencephalography network analysis

Brain connectivity has emerged as a neuronal analysis tool widely used to explore brain functions and supply relevant information in the study of the cognitive processes. However, current methodologies used to assess brain connectivity are not always exact and as a result, possible spurious connections may appear. Moreover, measuring the connection between all possible pairs of EEG-channels leads to high dimensional matrices with either redundant or irrelevant information. To avoid problems in connectivity analysis and issues of high computational cost, a selection stage of the most significant connections can be implemented. Nevertheless, there is not a standard method yet to extract connections and the definition of significant connections may vary accordingly with the object of study. Therefore, to develop an accurate methodology, information inherent to each specific problem should be included. In this work, three different tools are presented, that execute the extraction of significant connections considering the experimental scenario. The first tool, tested on a BCI dataset, finds the set of connections that best discriminate two MI classes. Consequently, a kernel-based methodology of feature selection is used to rank each connection by its contribution in the classes discrimination. Finally, the significant connections will be the smaller set that achieves the best classification accuracy. The second methodology is used in a study of the significant connectivity patterns in attention networks. To this end, the connectivity of two classes (target and non-target) in an oddball paradigm experiment is extracted. Here, the significant connections are selected as the ones that differ the most, statistically speaking, between target and non-target. Finally, in a study of the recovery of a subject with aphasia, differences in connectivity, related to improvements produced by therapy were found. In this study, connections that change through the sessions of treatment at the level of amplitude and structure were extracted. Also, a set of significant connections that changed increasingly between the sessions was selected. For all the proposed methodologies, the brain connectivity is computed over EEG signals and the extraction of the significant connections is based on information inherent to the data or the experiment. In general, the selection of connections allows the considerable reduction of connectivity characteristics, this facilitates the physiological interpretation of the experiments and can improve the performance and computational cost of the systems that use these features

• Autores: Hurtado Rincón, Juana Valeria
• Fecha de publicación: 2018
• Institución: Universidad Nacional de Colombia
• Repositorio: Universidad Nacional de Colombia
• Idioma: spa
• OAI Identifier: oai:repositorio.unal.edu.co:unal/64168
• Acceso en línea: https://repositorio.unal.edu.co/handle/unal/64168; http://bdigital.unal.edu.co/65000/
• Palabra clave: 62 Ingeniería y operaciones afines / Engineering; Brain connectivity; Electroencephalography; Significant connections; sdffsdf; Conectividad cerebral; Electroencefalografía; Conexiones significativas
• Rights: openAccess
• License: Atribución-NoComercial 4.0 Internacional