Analysis of Resting-State fMRI using Pearson-VII Mixture Modeling
Access rightsWorldwide access
MetadataShow full item record
In the past two decades, functional magnetic resonance imaging (fMRI) has been widely used to research and characterize neural activity in the brain based on measuring the hemodynamic response correlated to neuronal ﬁring. A main goal of fMRI analysis is to characterize functional connectivity – correlation in activation pattern – between diﬀerent regions of the brain for diﬀerent task- or disease-related events. Here, we present a novel, data-driven method for identifying functionally connected regions of the brain using a Pearson-VII mixture model learning algorithm (p7-means). This method is complementary to independent component analysis (ICA), which derives underlying source signals, rather than probabilistic distributions, to characterize brain components. The p7-means algorithm is powerful in its ability to model a range of leptokurtic to Gaussian components, which makes it robust in identifying core functional components in noisy images. Additionally, p7-means has the advantage of learning the number of components from the data set, rather than returning a ﬁxed number of components matching dimensionality. We apply the algorithm to resting-state monkey fMRI and compare the discovered components to those found in ICA. Correlational analysis shows consistent activation components between the two methods, although p7-means appears to result in more spatially localized groups.