Diffusion MRI of the Frontal Cortex

Previously, we reported cortical thickness changes in the right (trained) dlPFC in the region of Brodmann's area 9 [Sale et al., 2017]. Top‐down control of motor output, for error‐correction, is predominantly associated with activation of the dlPFC, as well as the caudate nucleus [Chevrier et al., 2007; Kübler et al., 2006]. The dlPFC is thought to connect strongly with premotor areas, including the supplementary motor area (SMA; [Lu et al., 1994])—a region which displayed changes in fMRI, cortical thickness, and TMS measures after training in the present cohort [Sale et al., 2017]. For these reasons, we measured FA in connections between the middle frontal lobe and SMA, and the middle frontal lobe and caudate nucleus, with diffusion tractography.

Probabilistic whole‐brain tractography was performed on constrained spherical deconvolution images using MRTrix 3.0. For each dataset, 20 million streamlines were generated using anatomically constrained tractography (iFOD2), with dynamic seeding and backtrack flags, but otherwise default settings. For each hemisphere, we sampled FA for tracts connecting (A) the middle frontal gyrus and caudate nucleus and (B) the middle frontal gyrus and SMA. Functional MRI‐driven tractography was not possible for this subanalysis due to practical limitations regarding scan time, and so a standard parcellation‐based approach was used. Labels used were from the single participant T1 templates (see above) to avoid temporal bias. The SMA label included both pre‐SMA and SMA‐proper. Sampling was weighted according to the output of MRTrix SIFT2 [Smith et al., 2015]. Difference in FA (post‐training–pre‐training) values violated assumptions of normality and were non‐symmetrical. As such, a binomial test was conducted to calculate for an over‐representation of FA increases (H₀: π ≤ 0.5) with Holm‐Bonferroni correction for four multiple comparisons (α < 0.05 FWE).