Data-driven fMRI group classification using connected components and Gaussian process classifiers

Functional magnetic resonance imaging (fMRI) is a popular tool for studying brain activity due to its non-invasiveness. Convention ally an expected response needs to be available for correlating with fMRI time series in model-driven analysis, which limits experimental paradigms to blocked and event-related designs. To study neuronal responses due to slow physiological changes, such as after a glucose challenge or a drug administration, for which the expected response is unavailable, we had proposed a data-driven method: connected component analysis. In this paper, a novel group classification method is proposed by using both connected components and Gaussian process classifiers. The results demonstrate that the method is able to differentiate insulin resistant volunteers from insulin sensitive volunteers by their neuronal response to glucose ingestion with an accuracy of 77%.