The Autism Brain Imaging Data Exchange (ABIDE): Analytical Approaches and Initial Results

Background:   The Autism Brain Imaging Data Exchange (ABIDE) consortium is a grassroots initiative aggregating and openly sharing to the scientific community over 1100 existing resting state fMRI (R-fMRI) data of individuals with Autism Spectrum Disorders (ASD) and age-matched typical controls (TC). This unprecedented neuroimaging dataset allows testing and generating new hypotheses in ASD. 

Objectives:   To demonstrate the feasibility of ABIDE to promote progress in understanding the neurobiology of ASD through discovery science. As a first step, given the increasing convergence in the ASD field on dysconnections among brain regions rather than local abnormalities, we focused on R-fMRI data which are particularly suitable for examining intrinsic functional connectivity (iFC). 

Methods:   Following examination and summary of the phenotypic characteristics of the sample (1112 datasets; 539 with ASD and 573 TC; 7-64 years), we carried out 1) full-brain iFC analyses for both structural and functional parcellation schemes (i.e., structural: Harvard-Oxford Atlas (HOA; Kennedy et al., 1998), functional: Craddock et al., 2012 [Crad-200]); 2) regional voxel-wise measures of iFC including regional homogeneity, voxel mirrored homotopic connectivity (VMHC), seed based iFC of the default network (DN), and fractional amplitude of low frequency fluctuations.  Group analyses, limited to males, accounted for age, FIQ, site, and mean framewise displacement (FD) correcting for voxel-wise multiple comparisons with Gaussian random field theory (Z > 2.3 and cluster-level p < 0.05). For whole brain parcellation-wise correlation analyses, corrections for false discovery rates were applied. 

Results:  Sample demographics reflect the current status of the neuroimaging field. Despite the lack of a priori coordination, most sites used standardized phenotyping applying the Autism Diagnostic Observation Schedule and the Autism Diagnostic Interview-Revised. We also identified factors that vary across studies, to guide future efforts to increase standardization. For example, DSM-IV-TR diagnoses were provided by 80% of sites; consistent with previous multi-site findings (Lord et al., 2012), sites varied markedly in DSM-IV-TR subtype distributions. While whole brain analyses revealed both hypo- and hyper-connectivity in ASD, hypo-connectivity dominated. Consistent with prior work, in ASD we found reduced iFC involving key DN nodes such as posterior cingulate cortex and dorsomedial prefrontal cortex, and reduced VMHC in sensorimotor cortex. Regional analyses highlighted ASD-related abnormalities in the thalamus, caudate, and insula. 

Conclusions:  The feasibility of establishing the ABIDE dataset reflects the rapid adoption of R-fMRI approaches in neuroimaging of ASD along with the benefits of standardized diagnostic assessment protocols. Evidence from this initial survey of the unprecedented ABIDE R-fMRI data provides demonstrations of both replication and novel discovery. It also demonstrates the vast information latent in any single functional imaging dataset, and the extraordinary statistical power available from combining datasets across investigators, labs, and countries. By pooling multiple international datasets, the ABIDE sample allows for replication, secondary analyses and discovery efforts, and is expected to accelerate the pace of discovery for the next generation of ASD studies.