The impact of in-scanner head motion on structural connectivity derived from diffusion MRI

Abstract

Multiple studies have shown that data quality is a critical confound in the construction of brain networks derived from functional MRI. This problem is particularly relevant for studies of human brain development where important variables (such as participant age) are correlated with data quality. Nevertheless, the impact of head motion on estimates of structural connectivity derived from diffusion tractography methods remains poorly characterized. Here, we evaluated the impact of in-scanner head motion on structural connectivity using a sample of 949 participants (ages 8-23 years old) who passed a rigorous quality assessment protocol for diffusion magnetic resonance imaging (dMRI) acquired as part of the Philadelphia Neurodevelopmental Cohort. Structural brain networks were constructed for each participant using both deterministic and probabilistic tractography. We hypothesized that subtle variation in head motion would systematically bias estimates of structural connectivity and confound developmental inference, as observed in previous studies of functional connectivity. Even following quality assurance and retrospective correction for head motion, eddy currents, and field distortions, in-scanner head motion significantly impacted the strength of structural connectivity in a consistency- and length-dependent manner. Specifically, increased head motion was associated with reduced estimates of structural connectivity for network edges with high inter-subject consistency, which included both short- and long-range connections. In contrast, motion inflated estimates of structural connectivity for low-consistency network edges that were primarily shorter-range. Finally, we demonstrate that age-related differences in head motion can both inflate and obscure developmental inferences on structural connectivity. Taken together, these data delineate the systematic impact of head motion on structural connectivity, and provide a critical context for identifying motion-related confounds in studies of structural brain network development.

Keywords: Artifact; Confound; DTI; Development; Motion; Structural connectivity; dMRI.

Figure 1. Connectome construction

For each subject (n=949, ages 8-23 years), the T1 image was processed using FreeSurfer and parcellated into 233 cortical and subcortical network nodes on a subject-specific basis. A ball-and-stick diffusion model was fit to each subject’s DTI data and probabilistic tractography was run with FSL probtrackx, initiating 1,000 streamlines in each seed voxel identified at the gray-white boundary for each node. Edge weights in 233×233 symmetric connectivity matrices derived from probabilistic tractography were defined by the number of streamlines connecting a node pair. Alternatively, the diffusion tensor was fit to the DTI data and deterministic streamline tractography was used to create a symmetric connectivity matrix (233×233), where the primary edge weight was defined by calculating the mean fractional anisotropy (FA) along streamlines connecting a node pair. Connection length was quantified by the mean length of streamlines connecting a node pair.

Figure 2. Motion effects on structural connectivity are driven by inter-subject edge consistency and streamline length

The effect of in-scanner head motion on structural connectivity was estimated using a partial correlation for each network edge while controlling for age, age², and sex. (A) When edge weights were defined by the number of probabilistic streamlines connecting a node pair, 12% of all network edges were significantly impacted by motion. (B) The direction and strength of motion effects were significantly correlated with inter-subject edge consistency (r=-0.35) and with mean streamline length (r=-0.21; see Supplementary Figure 1). (C) Inter-subject edge consistency exhibited a quadratic relationship with mean streamline length. Head motion significantly enhanced the strength of relatively short-range, low-consistency network edges. Further, head motion diminished the strength of relatively high-consistency network edges, which included both short- and long-range connections. (D) When edge weights were defined by the average FA along deterministic streamlines connecting a node pair, 14% of all network edges were significantly impacted by motion. (E) The direction and strength of motion effects were significantly associated with inter-subject edge consistency (r=-0.50) and with mean streamline length (r=-0.48; see also Supplementary Figure 1). (F) For networks derived from deterministic tractography, inter-subject edge consistency exhibited a parabolic relationship with mean streamline length. In agreement with results from probabilistic tractography, head motion significantly enhanced the strength of relatively short-range, low-consistency network edges, and diminished the strength of relatively long-range, high-consistency network edges. All statistical inferences were adjusted for multiple comparisons using FDR (Q < 0.05). The significance of all third-level correlations was evaluated using 10,000 permutations (permutation-based p < 0.0001). Black line in panels C and F represents the best fit from a general additive model with a penalized spline.

Figure 3. Head motion systematically impacts structural connectivity across consistency-based thresholds at the level of network edges, nodes, and total network strength

Motion effects on probabilistic edge strength, node strength, and total network strength were assessed across a range of consistency-based thresholds (ten thresholds, 0-90^th percentile inter-subject edge consistency). (A) The percentage of edges significantly impacted by head motion increased monotonically across consistency-based thresholds (12-32%). (B) After eliminating all edges with inter-subject consistency below the 50^th percentile, head motion significantly diminished the strength of 84% nodes, with particularly strong effects observed in middle frontal gyrus, precuneus, and cingulate cortex. (C) While the effect was stable across consistency-based thresholds, head motion significantly diminished total network strength at each threshold. Motion effects on deterministic edge strength, node strength, and total network strength were assessed across ten consistency-based thresholds (0-90% deterministic inter-subject edge consistency). (D) The percentage of deterministic network edges significantly impacted by head motion increased monotonically across consistency-based thresholds (14-62%). (E) After eliminating edges that existed in less than 50% of participant connection matrices, head motion significantly diminished the strength of 89% nodes, with particularly strong effects observed in the precuneus and medial brain regions including the anterior and posterior cingulate. (F) Head motion also significantly diminished total network strength across all consistency-based thresholds, particularly at more stringent thresholds. These results suggest that global strength normalization approaches may be confounded by individual differences in head motion during acquisition. All statistical inferences were adjusted for multiple comparisons using FDR (Q < 0.05). Black bars correspond to the standard deviation of 100 bootstrapped samples encompassing 80% of the dataset (n=760).

Figure 4. Observed age effects on structural connectivity are both inflated and obscured when age-related differences in head motion are not accounted for

All subjects included in this study passed rigorous manual quality assurance, retaining a sample of relatively high-quality, low-motion dMRI datasets. (A) Despite this, age-related differences in head motion were still observed: younger participants tended to move significantly more than older participants. (B) Mediation analyses across all network edges showing significant age effects demonstrated that observed age effects on structural connectivity were often inflated or obscured when head motion was not accounted for. This schematic illustrates how positive mediation effects can reflect inflated positive age effects or obscured negative age effects, where in both cases motion decreases the strength of network edges that undergo significant age-related change. Similarly, negative mediation effects can reflect inflated negative age effects or obscured positive age effects, where in both cases motion increases the strength of network edges that undergo significant age-related change. (C) For brain networks derived from probabilistic tractography, significant age effects were observed in 26% of all network edges. This visualization highlights 7% of these edges where developmental effects were significantly mediated by age-related differences in head motion. Positive mediation effects were observed for edges where motion significantly reduced connectivity, while negative mediation effects were observed for edges where motion significantly increased connectivity. (D) Network connections exhibiting positive mediation effects had significantly higher inter-subject edge consistency compared to connections with significant negative mediation effects (permutation-based p < 0.0001). (E) For brain networks derived from deterministic tractography, significant age effects were observed in 7% of all network edges. This visualization highlights 51% of these edges where developmental effects were significantly mediated by age-related differences in head motion. Again, both significant positive and negative mediation effects were observed. (F) As seen in the probabilistic data, network connections with significant positive mediation effects had significantly higher inter-subject edge consistency compared to connections with significant negative mediation effects (permutation-based p < 0.0001). Red connections in 4c through 4e represent significant positive mediation results; blue connections represent significant negative mediation results.