Supplementary Fig. 2: Image processing and analysis workflow. | Nature Neuroscience

Supplementary Fig. 2: Image processing and analysis workflow.

From: Compulsivity and impulsivity traits linked to attenuated developmental frontostriatal myelination trajectories

Supplementary Fig. 2

(a) Overview of applied longitudinal qMRI preprocessing workflow using SPM (and CAT toolbox, r1318) to normalize in standardized space (within-subject: Ashburner & Ridgway, 2012, Front Neurosci; between-subject: geodesic shooting, Ashburner & Friston, 2009, NeuroImage; 2011, NeuroImage) and smoothing (via tissue-weighted smoothing: Draganski et al., 2011, NeuroImage) of myelin-sensitive magnetization transfer (MT) saturation maps. First, we performed symmetric diffeomorphic registration for longitudinal MRI (Ashburner & Ridgway, 2012, Front Neurosci), combining nonlinear diffeomorphic and rigid-body registration and a correction for intensity inhomogeneity artefacts. The registration model creates a midpoint image for each subject and the corresponding deformation fields for every individual scan. Second, we applied CAT segmentation to each subject’s midpoint image, which assumes every voxel to be drawn from an unknown mixture of gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) tissue classes. The applied segmentation algorithm performs a partial volume estimation (PVE) to account for mixed voxels with two tissue types. Third, nonlinear template generation and image registration to MNI space was performed using the individual midpoint GM and WM tissue maps and diffeomorphic registration using geodesic shooting (Ashburner & Friston, 2011, NeuroImage). Quantitative MT maps from all time-points were normalized to the MNI space using (within- and between-subjects) transformations obtained in previous steps. Fourth, since our study aimed at local statistical analysis of voxel-based trajectories of quantitative MT parameters within GM and WM tissue classes, the normalized MT maps were smoothed using previously established (Draganski et al., 2011, NeuroImage) tissue-weighted-smoothing with a Gaussian kernel of 6 mm full width at half maximum (FWHM). Fifth, in order to avoid biasing results by image artefacts (for example due to movement, segmentation or normalization errors), the MPM maps were carefully checked manually before and after longitudinal registration by an expert. Additionally, the data was quality checked using a during scan motion proxy (Castella et al., 2018, Magn Reson Med; cf. Supplementary Fig. 8) and statistical covariance-based sample inhomogeneity measures (as implemented in the CAT toolbox) to exclude subjects with extremal overall deviation of quantitative values due to acquisition or processing artefacts. All processed quantitative MT and morphometric maps for subsequent longitudinal modelling steps were obtained using the above steps. (b) Schematic illustration of longitudinal voxel-based qMRI analysis in relation to developmental effects and compulsivity/impulsivity dimensions using SPM Sandwich Estimator toolbox r1.2.10 (SwE, Guillaume et al., 2014, NeuroImage). For more compact visualization of results, projections of template-space statistical (non-) parametric z-maps were performed on surface reconstruction of the same voxel-based template (gray and white matter tissue classes) used for normalization. (c) Comparing SwE toolbox output for parametric and non-parametric results for effect of developmental growth of MT within cortical regions (main results Fig. 1). Top row shows parametric inference correcting for multiple comparisons using voxelwise/peak-level FDR (p < .05, statistical Z-maps from one-sided Wald-test for MT increase with time/visits or mean age of subject). Middle row illustrates same FDR corrected results obtained using SwE toolbox under wild bootstrapping with 999 straps. Bottom row applies clusterwise FWE correction extent threshold k = 1147 (p = .05) with initial cluster forming threshold .001. The differences of FWE cluster and FDR voxelwise correction are expected, generally results were found to be similar in our ~500 scan sample suggesting validity of parametric inference scheme. SwE TFCE-based inference was found to be more sensitive than FWE (not shown).

Back to article page