On the Question of Brain Overgrowth in ASD: An in Depth Methodological Analysis Using the Large Abide Dataset

Background:  Neuroimaging findings of autism spectrum disorder (ASD) have produced inconsistent results even on global measures such as total intracranial volume (TIV). In comparison to age matched typically developing controls (TDC), studies suggest lower TIV in ASD at birth and overgrowth at early childhood; but there is disagreement at what age of onset this overgrowth occurs. 

Objectives:  We test the influence of site and preprocessing and analysis toolboxes (PAST) on ASD versus TDC TIV, gray matter volume (GMV) and white matter volume (WMV) differences. We also investigate if TIV growth curve is sensitive to PAST and provide evidence for methodological discrepancies. 

Methods:  Structural MRIs were obtained from the Autism and Brain Imaging Data Exchange^. (ABIDE). We preprocessed 417 ASDs (average age ± std: 17.7 ± 8.2 years) and 459 TDCs (17.7 ± 7.9 years) using SPM, FSL and FreeSurfer (FS). After accounting for age, IQ, sex and site for each and all 15 ABIDE sites we performed two sample t-tests on residual brain volumes. Growth curves were computed for all ABIDE subjects within the age window of 6 to 40 years using local polynomial regression fitting.

Results:  In Figure 1A, B and C we present boxplots for TIV, GMV and WMV computed with SPM (in red), FSL (in green) and FS (in blue) along with ASD vs. TDC t-test p-values for all 15 sites. When TIVs from all sites were compared, only SPM showed significantly higher TIV in ASD, FSL showed lower TIV in ASD and FS higher TIV in ASD (insignificant). Site wise comparisons were inconsistent. Only one site (Caltech) showed significantly higher TIV volume in ASD with all three methods. Out of the other fourteen sites, SPM showed higher ASD TIV volume in 10 sites and FSL and FS only in 6 sites (not same). For GMV and WMV, SPM showed higher values in ASD for most sites, but FS and FSL results were inconsistent. Growth curves (Figure 1D) indicate lower ASD TIV at birth and later overgrowth across all three methods but with differing onset of overgrowth (SPM: 7, FSL: 15 and FS: 10 years with possible degeneration around 30 years). In Figure 2A we present the standard deviations of TIV indicating generally higher TIV variance in ASD and FS. In Figure 2B SPM vs. FS and SPM vs. FSL TIV scatter plots for all ABIDE subjects indicate that compared to SPM, FS can either under or overestimate TIV and FSL under estimates TIV in most subjects. In Figure 2C and 2D we overlay GM and CSF maps of SPM and FSL. Although most GM regions overlap, CSF estimates have many non-overlapping regions.

Conclusions:  By applying three different volume computation methods to a very large dataset we show that in addition to inter-site heterogeneity of ASD brain volume, poor inter-method reliability may also contribute toward inconsistent results. Note: subjects used in this study were older than 6 years and hence we could not test the popular early brain growth hypothesis.