Group Differences in Head Motion May Confound Anatomical Connectivity Findings from DWI

Background:  Common findings in diffusion-weighted imaging (DWI) studies of children and adolescents with autism spectrum disorder (ASD) include reduced fractional anisotropy (FA), increased mean diffusivity (MD), and increased radial diffusivity (RD) of white matter tracts. Such effects have been reported by multiple groups particularly in frontal and temporal lobes and in the major association fasciculi. However, recent studies suggest that head motion may alter diffusion measures and result in spurious findings of group differences (Ling et al. 2012, Koldewyn et al. 2014, Yendiki et al. 2013), calling these results into question.

Objectives:  To determine whether careful matching for head motion in DWI would alter findings of group differences in children and adolescents with ASD.

Methods:  Diffusion weighted MRI was collected from 54 ASD and 44 typically developing (TD) participants ages 7-17 years. In a multi-stage process, groups were matched at increasing levels of stringency based on qualitative and/or quantitative assessment of head motion. Qualitative assessment included visual inspection for (i) slice-wise signal dropout (ii) image noise and (iii) shifts of head placement between acquisition of diffusion volumes (sensitization directions).    

Quantitative assessment included four motion measures (average inter-volume translation, average rotation, proportion of slices affected by signal dropout, severity of signal dropout) as described by Yendiki et al. (2013). At each matching stage, groups were compared on FA, MD, RD, and AD (axial diffusivity) using two approaches: Tract Based Spatial statistics (TBSS) and probabilistic tractography.

Results:  When quality screening was not applied (n_ASD=54, n_TD=44) the ASD (compared to the TD) group showed higher RD diffusely throughout the right cerebral hemisphere, with modest effect sizes (Cohen’s d≈.27) and no other significant differences. We then excluded participants with any visible artifacts in the raw DWI images (stringent quality group: n_ASD=30, n_TD=30), after which no significant differences were found between groups on any diffusion measures. However, assessment of quantitative measures of motion showed residual differences in average motion between groups. When these carefully screened groups were further matched on the four quantitative measures of motion (n_ASD=27, n_TD=22), significantly higher RD was found in the right forceps major and splenium of the ASD group, with large effect sizes (d≈.83).

In preliminary analyses, we also performed probabilistic tractography on white matter projection paths, specifically the cortico-spinal tract and striato-frontal pathways, detecting significant group differences for cortico-spinal (AD) and striato-motor (MD, RD) tracts in full samples, but no significant group differences in stringently quality-controlled and matched groups.

Conclusions:  Our results are partially consistent with those reported by Koldewyn et al. (2014), suggesting that group differences in head motion can have substantial effects on DWI findings. However, in contrast to this earlier study, we found that optimal group matching re-instated some between-group findings not seen in less tightly matched, but stringently quality-controlled data. These differences may reflect a subtle interplay between data quality, matching, and reduced statistical power in more selective subsamples. Given inadequate motion procedures in many previous DWI studies of ASD, the existing literature may have to be revisited with caution.