Objectives: We sought to determine if and how spatial correlation and variance heterogeneity affect estimated individual and group differences in corpus callosum WMM in autism.
Methods: We examined the voxel-wise mean and variance of fractional anisotropy (FA) and a novel full tensor measure (MV) in the genu and splenium in a sample of N=80 children and young adults (mean age 14.1 years) with ASD and N=40 typically-developing matched controls (mean age 15.5 years). We compared results from models that assumed intra-subject, inter-subject and inter-group variances to be equal or unequal across subjects and between groups. All models accounted for inter-voxel correlation in each subject.
Results: After accounting for group mean differences, we found that inter-subject variance was significantly high in splenium FA (χ2=6.0, df=1, p=0.01 and genu MV (χ2=17.4, df=6, p<0.01). Second, we found that autism subjects had significantly greater intra-subject FA variance in these regions (χ2=16.7, df=1, p<0.0001; χ2= 44.2, df=1, p<0.0001) and significantly greater MV variance as well (χ2= 263.2, df=6, p<0.0001; χ2=117.2, df=6, p<0.0001) in autism. Last, we observed substantially more coherent clustering of FA, MV and their group differences by accounting for inter-voxel correlation compared to those derived by a massively univariate approach.
Conclusions: Our results suggest that WMM in the genu and splenium of the adolescent male with high-functioning autism is more heterogeneous than that of his typically developing counterpart. There is a high likelihood of decreased consistency of directional diffusion coherence in the corpus callosum of an individual with autism and between individuals with autism, perhaps due to genetic and/or epigenetic dysregulation in brain development. In addition to the first-order (mean) decreases in directional diffusion coherence between groups reported previously, these second-order (variance) deviations suggest that a variety of disruptions of fiber organization may affect the quality of inter-hemispheric information transfer and perhaps language and social functioning. Our novel method demonstrated that (1) when testing for group differences in WMM, particularly in the corpus callosum and possibly other brain regions, one should acknowledge the correlation and increased variance of voxel-wise tensor measures in autism relative to measures in healthy populations; and (2) the detection of a regional group difference at a single voxel that acknowledges regional voxel-wise correlations outperforms widely-used approaches that ignore them.