A functional connectivity primer

Abstract

The use of in-vivo imaging methods to study connectivity in the brain has grown dramatically over the past several years. While a large part of this growth is due to the availability of diffusion tensor imaging (DTI), methods for examining connectivity using widely available BOLD contrast data have also been growing. The umbrella term for the latter, functional connectivity, was defined by Friston et al. (1993) as temporal correlation between spatially remote neurophysiological events. This definition captures the essentially correlative nature of these methods regions are considered functionally connected if their activity is in some way correlated, regardless of the mechanism underlying the correlation. As contrasted with effective connectivity (the influence that one neural system exerts over another either directly or indirectly), functional connectivity doesn't necessarily imply a physical pathway, and potentially includes patterns of connectivity that are entirely mediated by the common influence of some external event on distant neural areas. Much of the study of functional connectivity (FC) is carried out by examining inter- regional correlations in resting BOLD data, an approach that is traced to a study by Biswal et al. (1995; see Rogers et al., 2008 for a brief note on earlier related methods), who observed correlations between activity in left and right somatosensory cortex during resting BOLD. Temporal correlations in resting data are of special interest because they are not easily explained by externally imposed task demands (although they may be influenced by endogenously driven behavior at rest). At the same time, a growing literature on neural activity during the kinds of passive experimental conditions ordinarily used as control or rest conditions during fMRI suggests that resting BOLD is more than just the absence of cognitively evoked activity. By contrast, connectivity measures from data collected during task performance risk discovering trivial associations of little novel interest there is an obvious reason why regions in the left and right motor cortex should be correlated during bimanual finger tapping, and a functional connectivity analysis sheds no new light on these processes. A major practical advantage of connectivity studies carried out in resting data is that the same data may be used repeatedly. At this point, a great deal of resting BOLD data is available either publicly or through local repositories, and many questions can be addressed without the collection of new data. Further, Fair et al. (2006) have demonstrated that rest periods from blocked-design BOLD studies yield similar results to what can be obtained from simple resting BOLD data, which opens up the possibility of drawing on an even larger store of existing data. Understanding spatial patterns of intercorrelation may be important for another reason. Fox et al. (2006) have shown that some of the coherent signal in resting BOLD contributes roughly linearly to task-evoked BOLD. In this way, better understanding of the task- independent component of the signal can lead to markedly better sensitivity to detect task- evoked activation. The purpose of this article is to convey the essential details of the most common approaches to studying functional connectivity in BOLD data, mostly resting BOLD. I first describe typical processing steps involved in these analyses, then describe a few variants of the basic method, and finally describe available software tools for carrying out these analyses.