Abstract

Functional magnetic resonance imaging (fMRI) has great potential for developing theories of functional connectivity in large-scale neural networks. This is due to its ability to localize neural activity simultaneously in the entire brain. Standard univariate analysis techniques, however, provide only a limited perspective on fMRI data. By definition, univariate analyses do not allow for direct comparisons of activity between two or more regions. In contrast, bivariate analyses can provide information about a network of activity associated with a particular seed or region of interest (ROI). We propose a bivariate method for analyzing fMRI data using coherence, a measure of the strength of the linear relation between two signals at every frequency. Coherence also has an additional advantage in that it obviates any reliance on a model of the hemodynamic response. Using simulated and acquired event-related fMRI data, we evaluated the effectiveness of coherence as an measure of functional connectivity between regions. We developed a method for calculating condition-specific coherence maps of the entire brain volume from a single seed-voxel or ROI. When a seed-ROI is chosen based on task relevance, the map of voxels coherent with the seed indicates regions that are not only task relevant, but also functionally coupled to the ROI. Our results show that bivariate coherence maps are an important complement to standard univariate techniques for investigating theories of functional connectivity.

From: Sun, F. T., L. M. Miller, et al. (2002). Using coherence of inter-regional fMRI time-series as an index of functional connectivity. Society for Neuroscience, Orlando.