Short communication
Network homogeneity reveals decreased integrity of default-mode network in ADHD

https://doi.org/10.1016/j.jneumeth.2007.11.031Get rights and content

Abstract

Examination of spontaneous intrinsic brain activity is drawing increasing interest, thus methods for such analyses are rapidly evolving. Here we describe a novel measure, “network homogeneity”, that allows for assessment of cohesiveness within a specified functional network, and apply it to resting-state fMRI data from adult ADHD and control participants. We examined the default mode network, a medial-wall based network characterized by high baseline activity that decreases during attention-demanding cognitive tasks. We found reduced network homogeneity within the default mode network in ADHD subjects compared to age-matched controls, particularly between the precuneus and other default mode network regions. This confirms previously published results using seed-based functional connectivity measures, and provides further evidence that altered precuneus connectivity is involved in the neuropathology of ADHD. Network homogeneity provides a potential alternative method for assessing functional connectivity of specific large-scale networks in clinical populations.

Introduction

The advent of new methods for analyzing functional neuroimaging data in the resting state has enabled the investigation of previously overlooked aspects of intrinsic network organization. In particular, investigators have identified spontaneous coherent fluctuations in functionally distinct networks even in the absence of specific cognitive instruction (De Luca et al., 2006, Fox et al., 2006, Vincent et al., 2006). This “cognitively unbiased” approach appears to be particularly relevant for the study of psychopathological populations, with several recent reports noting disruptions in such intrinsic organization (Greicius et al., 2004, Garrity et al., 2007).

Assessment of resting brain networks can be accomplished using several recently developed methods, although two are most widely employed. These are independent components analysis (ICA) and region-of-interest (ROI) seed-based correlation approaches, each of which has strengths and shortcomings. ICA is a model-free approach whereby a two-dimensional (time points x voxels) data matrix is decomposed into a set of independent timeseries and consequently associated spatial maps which describe the temporal and spatial characteristics of the underlying signals (components) (Beckmann et al., 2005). While ICA has the power to estimate largely overlapping spatial processes, there is no clear consensus as to how best to compare components across subjects and/or between groups (Fox and Raichle, 2007) (see (Beckmann and Smith, 2004, Calhoun et al., 2005) for recent advances). Seed-based approaches involve using correlation or regression analyses to determine the temporal coherence between the timeseries for a particular voxel or ROI and the timeseries of all other voxels in the brain in order to identify temporally coherent “functionally connected” networks (Biswal et al., 1995). This flexible approach is limited by the requirement for an a priori ROI seed region, which is typically selected based upon prior findings or theoretical models. Unfortunately, determining optimal ROI seed placement within a network is non-trivial, as brain circuits can encompass a large number of brain regions consisting of multiple functionally differentiable sub-regions (Margulies et al., 2007). The decision to place the ROI seed in one sub-region within this network can be somewhat arbitrary, with no way to determine which placement might be “best”. The selection and precise placement of ROI seeds can therefore have considerable impact on the patterns of functional connectivity observed. Consequently, decisions made early in the analytic process concerning seed placement could potentially result in lack of appreciation of abnormalities in functional connectivity in a particular clinical population. For the same reason, if no a priori prediction exists for a particular region or ROI, potential abnormalities in the associated networks could be missed.

Alternative approaches are currently being developed to address some of the limitations of the prevailing techniques. In order to overcome the difficulty of sorting ICA components, Greicius et al. have developed a template-matching procedure which identifies a network of interest by its goodness-of-fit to a pre-specified template mask (Greicius et al., 2004). Wang et al. have developed a parcellation method, which divides the whole brain into several regions to identify abnormal connectivity by comparing correlation coefficients of each pair of regions between groups (Wang et al., 2007). A similar method of computing regionwise correlation matrices has recently been proposed (Fair et al., 2007). All of these newer approaches share the advantage of simplifying between-group comparisons of resting-state fMRI data. Another approach, “regional homogeneity”, measures the similarity of the timeseries of a given voxel to those of its nearest neighbors (Zang et al., 2004). This method has been used to demonstrate abnormalities in a fronto-striatal-cerebellar functional network in attention-deficit/hyperactivity disorder (ADHD) (Cao et al., 2006). While informative, this “regional” method is only sensitive to the temporal synchrony of the BOLD signal within the 26 neighboring voxels for any given voxel, and thus is poorly suited for studying long-range connectivity.

When assessing network integrity in clinical populations, one potentially informative approach would be to provide an unbiased survey of a distributed network of interest, looking for regions exhibiting pathology-related decreases in network coherence. Here we present a novel measure for this purpose, which we term “network homogeneity”. This is a voxel-wise measure that provides an assessment of a voxel's correlation with all other voxels within a given network of interest. This measure is defined as the mean correlation of any given voxel's timeseries with the timeseries of every other voxel in the network. Brain regions exhibiting compromises in network homogeneity in association with a particular disorder or pathological process can be identified by our method, which enables between-group comparisons. A primary advantage of this approach is that it provides an unbiased survey of a given network, so that group differences may be identified without the need for a priori knowledge of where in the network abnormalities might be.

The most prominent network in the clinical neuroscience literature on spontaneous intrinsic brain activity is the default mode network (DMN). The DMN comprises medial (medial prefrontal cortex, posterior cingulate/precuneus) and lateral (posterior parietal) brain regions that routinely exhibit coherent decreases in activity during attention-demanding cognitive tasks (Raichle et al., 2001). Attentional lapses have been found to occur shortly after periods of decreased deactivation of posterior DMN regions (Weissman et al., 2006). ADHD is a heterogeneous developmental condition with multiple potential loci of neural dysfunction. Though there are numerous theoretical reasons for suspecting DMN dysfunction in ADHD, little empirical work has been conducted in this area. In a recent review, DMN interference during task performance was suggested to be a potential underlying cause of performance variability in ADHD (Sonuga-Barke and Castellanos, 2007). In a recent study, examining resting state functional connectivity of the dorsal anterior cingulate cortex using a seed-based approach, we reported a secondary finding of decreased functional connectivity between the precuneus and other DMN regions in adults with ADHD (Castellanos et al., in press). This seed-based approach consisted of extracting the timeseries of specific ROIs selected from previous work (Weissman et al., 2006) and using these as regressors to produce maps of all positively and negatively predicted voxels for each regressor (Margulies et al., 2007). This analysis revealed decreased functional connectivity between the anterior cingulate cortex and precuneus in ADHD subjects. Secondary analyses using the precuneus as a starting point for seeding revealed decreases in connectivity between the precuneus and ventromedial cortex in ADHD participants. Though examination of the DMN was not the primary focus of that study, results suggested that more detailed examination of DMN integrity in ADHD was warranted. Here we apply our network homogeneity measure to the same dataset to demonstrate its efficacy as a complementary method to seed-based functional connectivity and ICA. Based on our previous findings and the functional associations of this network, we hypothesized that we would find abnormal DMN homogeneity in the ADHD group compared to the control group, and that regions of “disconnect” would confirm and extend those previously identified via seed-based approaches.

Section snippets

Participants

Twenty adults with ADHD were recruited from the New York University School of Medicine Adult ADHD Program, and 20 age-matched comparison subjects were recruited through local media advertisements. All prospective participants were screened with the Symptom Checklist-90-Revised (SCL-90-R) to exclude a broad range of psychiatric psychopathology (Deragotis, 1986). Exclusion criteria for both groups included: (1) lifetime history of psychotic, bipolar or substance use disorders, (2) current history

Results

The ADHD and control groups did not differ on global measures of DMN homogeneity (mean ADHD = 0.0345, mean Control = 0.0357, t(39) = 2.7, p > 0.01), suggesting against the possibility of diffuse compromises in network integrity. For voxel-wise group differences tested using a random effects model implemented in SPM2, the comparison Control > ADHD revealed ADHD-related reductions in network homogeneity in the posterior portion of the DMN centered in precuneus (see Fig. 2). The peak difference between the

Discussion

To provide an unbiased survey of within-network coherence capable of detecting specific loci of compromised connectivity, we have developed a novel measure termed “network homogeneity”. We applied this measure to resting state fMRI data to directly assess the integrity of the DMN in ADHD. This approach to characterizing DMN integrity revealed that compared to age-matched controls, subjects with ADHD showed decreased network homogeneity, particularly in the region of the precuneus. Importantly,

Acknowledgments

The authors would like to thank Dr. Andrew Kirsch for conducting clinical assessments. This work was supported by grants provided to FXC by the Stavros S. Niarchos Foundation, NIMH (5R21MH066393 and 5T32MH067763), and the Leon Lowenstein Foundation.

References (30)

  • B. Biswal et al.

    Functional connectivity in the motor cortex of resting human brain using echo-planar MRI

    Magn Reson Med

    (1995)
  • Q. Cao et al.

    Abnormal neural activity in children with attention deficit hyperactivity disorder: a resting-state functional magnetic resonance imaging study

    Neuroreport

    (2006)
  • Castellanos FX, Margulies DS, Kelly C, Uddin LQ, Ghaffari M, Kirsch A, Shaw D, Shehzad Z. Di Martino A, Biswal B,...
  • F.X. Castellanos et al.

    Quantitative brain magnetic resonance imaging in attention-deficit hyperactivity disorder

    Arch Gen Psychiatry

    (1996)
  • A.E. Cavanna et al.

    The precuneus: a review of its functional anatomy and behavioural correlates

    Brain

    (2006)
  • Cited by (368)

    View all citing articles on Scopus
    View full text