The basal ganglia are subcortical brain structures important for motor, cognitive, and emotional processing (Mink, 1996
). The consequences of basal ganglia pathology can be devastating, exemplified by the symptoms of degenerative basal ganglia disorders such as Parkinson's and Huntington's disease. Understanding the location and functional connectivity patterns of basal ganglia divisions would improve cognitive neuroscience investigations. Indeed, methods that could identify putative basal ganglia divisions are needed to test hypotheses about cortical-basal ganglia circuitry in typical development (Rubia et al., 2006
), healthy aging (Hedden and Gabrieli, 2004
), and disorders (e.g., Parkinson's disease, Huntington's disease, Tourette's syndrome) and are critical for region identification needed to develop more precise models of whole-brain connectivity (Butts, 2009
There are multiple levels of organization in the basal ganglia. Anatomically, the basal ganglia comprise five gray matter nuclei: the caudate, putamen, globus pallidus, substantia nigra, and subthalamic nucleus. The majority of projections from the cerebral cortex to the basal ganglia terminate in the caudate and putamen, collectively referred to as the striatum. Discrete cerebral cortical regions project to discrete striatal regions that then project, via the thalamus, back to those cortical regions (Alexander et al., 1986
). Within the striatum are compartments, termed patches and matrices, that have distinct neurochemical markers and receive projections from different cortical layers (Gerfen, 1989
; Graybiel, 1990
). Beyond basal ganglia nuclei that can be seen on structural MRI scans, more fine-grained divisions in human basal ganglia, though presumed to exist based on non-human primate and rodent studies, are difficult to identify with current neuroimaging methods.
While historically considered to be a motor structure, the basal ganglia receive cortical projections from all lobes of the cerebral cortex and contribute to both motor and non-motor processing (Mink, 1996
). Anatomical tracer studies in non-human primates (Alexander et al., 1986
; Middleton and Strick, 2000
; Haber, 2003
) have documented anatomical connections between the basal ganglia and many regions in the cerebral cortex, including lateral prefrontal, orbitofrontal, anterior cingulate, lateral parietal, motor, premotor, oculomotor, somatosensory, auditory association (superior temporal gyrus), and visual association (inferior temporal gyrus) cortex.
Resting-state functional connectivity MRI (rs-fcMRI) and diffusion tensor imaging (DTI) provide a means to assess functional and anatomical connectivity non-invasively in humans. It is important to note at the outset that these methods yield distinct information about brain connectivity. rs-fcMRI measures correlations in low-frequency (i.e., <0.1
Hz) spontaneous blood oxygenation level-dependent (BOLD) signal fluctuations (Fox et al., 2005
) and may reflect a history of co-activation between regions (Fair et al., 2007
; Dosenbach et al., 2008
). DTI measures the diffusion of water molecules, which is constrained by the presence of axons, particularly myelinated axons, and provides indices of white matter coherence used to create visualizations of white matter tracts. While there can be overlap in connectivity patterns identified using rs-fcMRI and DTI, functional connectivity has been documented in the absence of anatomical connectivity. For example, seeds placed in voxels corresponding to left and right retinotopic eccentric representations in primary visual cortex exhibit strong functional connectivity with rs-fcMRI, but are not anatomically connected (Vincent et al., 2007
). This observation suggests that functional connectivity should not be treated as a measure simply homologous to anatomical connectivity.
Despite fundamental differences in the types of information about brain connectivity that can be gleaned from rs-fcMRI and DTI, these methods converge with evidence from anatomical tracer studies examining cortical-basal ganglia connectivity, revealing significant connectivity between basal ganglia regions and frontal, parietal, and temporal regions. Using rs-fcMRI, dorsal and ventral caudate and putamen regions of interest (ROIs) were shown to have different patterns of functional connectivity with the cerebral cortex (Di Martino et al., 2008
; Harrison et al., 2009
). Similarly, large-scale cortical ROIs (e.g., prefrontal cortex, parieto–occipital cortex) were shown to have different patterns of partial correlations with the basal ganglia (Zhang et al., 2008
). DTI investigations have revealed different anatomical connectivity between basal ganglia divisions and large-scale frontal ROIs (e.g., prefrontal cortex, orbitomedial frontal cortex) (Lehericy et al., 2004
; Leh et al., 2007
; Draganski et al., 2008
). Across these methods, convergent findings regarding patterns of cortical-basal ganglia connectivity have emerged. For example, both rs-fcMRI and DTI respectively reveal functional and anatomical connectivity between dorsal caudate and lateral prefrontal cortex, ventral striatum and orbitofrontal cortex, and dorsal caudal putamen and motor and premotor cortex (Lehericy et al., 2004
; Leh et al., 2007
; Di Martino et al., 2008
; Draganski et al., 2008
; Harrison et al., 2009
Basal ganglia divisions have two properties that would facilitate identification with noninvasive neuroimaging methods: they have different patterns of connectivity with the cerebral cortex and they are spatially constrained (i.e., discrete) entities (Alexander et al., 1986
). Thus, it may be possible to identify basal ganglia divisions smaller than nuclei on the basis of their unique patterns of cortical-basal ganglia functional connectivity using rs-fcMRI and community detection algorithms, which are used to identify groupings in networks. rs-fcMRI is sensitive to changes in patterns of functional connectivity across adjacent, proximal (i.e., ~2
cm apart) cortical regions. For example, rs-fcMRI data contained abrupt transitions, consistent with boundaries between putative cortical areas, in the measured similarity of functional connectivity maps generated from seeds placed along a line between supramarginal and angular gyrus regions (Cohen et al., 2008
). Rather than simply measure along a single line, rs-fcMRI methods can also be used to sample from a larger structure (e.g., the basal ganglia). By calculating the similarity in whole-brain rs-fcMRI maps generated from each voxel in a structure, we can obtain a matrix of the pairwise similarity relationships between voxels. Similarity matrices can be used to bring recent developments in graph theory, the mathematical description of networks, to bear on our question of identifying divisions in the basal ganglia.
In graph theory parlance, a graph is composed of two elements: nodes, which represent the units of observation in a graph, and edges, which represent the pairwise relationships between nodes. We can thus view our similarity matrix as a network, with voxels as nodes and eta2
values, a measure of similarity, as edges. Community detection algorithms (e.g., modularity optimization [Newman, 2006
] used here) can be applied to cluster the nodes into highly interconnected communities, with relatively few edges between communities. In other words, these algorithms can be viewed as grouping voxels with similar correlation maps. Returning to our question of interest, these groupings can be examined to determine whether they reflect expected divisions within the basal ganglia. If (1) the anatomical loci of modularity optimization groupings is consistent with basal ganglia divisions identified from anatomical studies in non-human primates and rodents and (2) functional connectivity maps generated from the modularity optimization groupings are consistent with presumed patterns of cortical-basal ganglia connectivity, then we will consider these groupings to be putative basal ganglia divisions.
In this paper, we demonstrate that a novel approach to functional mapping that combines rs-fcMRI and modularity optimization analyses can reveal putative basal ganglia divisions in individuals. Our approach identifies putative basal ganglia divisions with reliable patterns of functional connectivity with an amount of data that can be acquired in a single, brief MRI session (i.e., one ~8-min structural scan and three ~5-min scans of relaxed fixation). Remarkably, these results appear to be robust at the individual subject-level.