Detecting resting-state networks of the human brain through the application of independent component analysis to neuroimaging data

Abstract

Independent component analysis (ICA) is an exploratory statistical technique used to extract latent variables from multivariate data. The latent variables, called components, are found by maximising the independence between them and so, a proxy for independence is required. Two popular approaches are to either minimise the mutual information between components, or maximise the non-Gaussianity between components. The technique of ICA is especially powerful when applied to the field of neuroimaging, where it is used to identify hidden signals that correspond to resting-state networks (RSNs). The term resting-state networks refers to brain regions which are connected due to their function. Hence, determining RSNs is a way of studying the functional connectivity that is present within the human brain. This dissertation aims to study the theoretical aspects of the ICA technique, and explore its application to functional magnetic resonance imaging (fMRI) data. A linear probabilistic ICA (PICA) model was applied to detect the RSNs in a human brain, where a discussion was provided on the means of classification.