The ADHD-200 Consortium: A Model to Advance the Translational Potential of Neuroimaging in Clinical Neuroscience

Neuropsychiatric imaging remains a pioneering frontier in modern medicine. Recent decades have witnessed marked advances in identifying biological correlates for a broad array of illnesses (Hillary et al., 2007; Ritsner, 2009; Linden and Thome, 2011; Shenton and Turetsky, 2011). However, our understanding of the underlying pathophysiology of neuropsychiatric illnesses remains insufficient (Ecker et al., 2010; Linden, 2012). Equally problematic, translational promises have yet to be delivered, as clinically useful biomarkers are rarely attained (Hyman, 2002; Nestler and Hyman, 2010). As such, psychiatry remains uniquely reliant upon a diagnostic and classification system derived from clusters of symptoms rather than etiology or neurobiology (Hyman, 2007; van Praag, 2008; Nesse and Stein, 2012). Recent works demonstrating the feasibility of predicting maturational and disease status from functional MRI and morphometric imaging data (Craddock et al., 2009; Dosenbach et al., 2010; Ecker et al., 2010) have rekindled hopes for the eventual development of imaging-based tools to inform clinicians in their efforts (Bullmore et al., 2009; Fox and Greicius, 2010; Bullmore, 2012; Klöppel et al., 2012; Michel and Murray, 2012). While these approaches are promising, substantial obstacles remain that can drastically hinder the pace of progress if left unaddressed (Kelly et al., 2012). In particular, the availability of largescale imaging data is of paramount importance to the advancement of human brain imaging in neuropsychiatry (Van Horn and Gazzaniga, 2002; Buckner, 2010; Yeo et al., 2011; Milham, 2012). Myriad hypotheses exist regarding the etiology and manifestations of pathologic processes in the brain. It is only through the acquisition of large-scale imaging data with appropriate phenotyping (Bilder et al., 2009a,b; Cohen et al., 2011) that these hypotheses can be properly evaluated. Simultaneously, such datasets are a prerequisite to the deployment of discovery science approaches, which have the potential to yield more precise and empirically grounded hypotheses. Unfortunately, datasets of the prescribed scale are unprecedented in the imaging community, and particularly challenging for psychiatric imaging given its burdens (e.g., extensive time and substantial costs of recruitment, psychiatric assessment, and phenotyping). Individuals affected by psychiatric illness, as well as children, are also prone to a higher frequency of data loss due to motion (Power et al., 2012; Satterthwaite et al., 2012; Van Dijk et al., 2012; Wilke, 2012) and inability to tolerate the scanner environment, which only exacerbate the difficulties. Fortunately, the 1000 Functional Connectomes Project (FCP) provided a model through which large-scale datasets can be obtained (Biswal et al., 2010; Milham, 2012). Specifically, the FCP pooled previously collected data from independent sites around the world, and demonstrated that discovery science could be performed on the aggregate sample. The FCP model of open sharing for the purposes of hypothesis testing and generation was not new, as a number of like minded efforts attempted sharing in the past (Van Horn et al., 2001; Marcus et al., 2007b; Weiner et al., 2012). Arguably, the FCP capitalized on the greater ease of sharing structural and resting state functional MRI datasets, whose methods are more amenable to sharing than taskbased datasets. In addition, it highlighted the increasing willingness of many laboratories to participate in open science. Still, the FCP’s success only represents an initial step in the implementation of open sharing in the imaging community as it only included non-clinical samples with phenotypes limited to age and sex. Building on this model, functional neuroimaging investigators working on Attention-Deficit Hyperactivity Disorder (ADHD) in three continents came together to form the ADHD-200 Consortium (see Acknowledgments for ADHD-200 Consortium details). The effort was to establish a large-scale, aggregate resting state fMRI dataset, along with accompanying anatomical and phenotypic data for children and adolescents with ADHD. The consortium publicly released 776 resting state fMRI and anatomical datasets collected at eight independent imaging sites on March 1, 2011 (Table 1). Included were 491 datasets obtained from typically developing individuals and 285 from children and adolescents diagnosed with ADHD, all between the ages of 7–21 years. The release was coordinated through the International Neuroimaging Data sharing Initiative (INDI), which makes use of the web infrastructure provided by Neuroimaging Informatics Tools and Resources Clearinghouse (NITRC) NITRC.org. Accompanying phenotypic information includes: diagnostic status, dimensional ADHD symptom measures, age, sex, intelligence quotient (IQ), and lifetime medication status. Additionally, preliminary quality control assessments (usable vs. questionable) based upon visual time-series inspection were included for all resting state fMRI scans. The ADHD-200 release data are stored and distributed in two ways: via NITRC Resources (NITRC-R) as The ADHD-200 Consortium: a model to advance the translational potential of neuroimaging in clinical neuroscience