Computational protocol: Increased functional connectivity between cortical hand areas and praxis network associated with training-related improvements in non-dominant hand precision drawing

Similar protocols

Protocol publication

[…] Structural and functional fMRI data were preprocessed and analyzed using fMRIB's Software Library (FSL v.5.0, http://www.fmrib.ox.ac.uk/fsl/) () and involved several steps: Non-brain structures were removed using BET. Head movement was reduced using MCFLIRT motion correction. EPI unwarping was performed to correct for distortions due to magnetic field inhomogeneities using FSL PRELUDE and FUGUE, using a separate fieldmap collected following the functional runs. The data were spatially smoothed using a Gaussian kernel of 6 mm FWHM. Slice timing correction was applied. For each data set, intensity normalization was applied using “grand mean scaling”, wherein each volume in the data set was scaled by the same factor to allow for valid cross-session and cross-subject statistics. A high-pass temporal filter with 200-second cut-off was applied to remove low-frequency artifacts. Independent component analysis (ICA) was conducted with MELODIC to denoise the data, following procedures detailed below. Field maps were used to apply B0 unwarping. Time series statistical analysis was carried out in FEAT v.6.00 using FILM with local autocorrelation correction (). Functional data were registered with the high-resolution structural image using boundary-based registration (), and resampled to 2 × 2 × 2 mm resolution using FLIRT; the participant images were then registered to standard images (Montreal Neurological Institute MNI-152) using FNIRT nonlinear registration ().MELODIC ICA was used to identify artifactual components for removal. Components were rejected if they clearly met one of the following criteria: (1) preponderance of suprathreshold voxels in non-brain areas, including ventricles or a “halo” outside the brain; (2) spin history effects, specifically activations that appear in alternating slices; (3) time course dominated by distinct temporal spikes; (4) components covering the majority of 1–2 slices, but not neighboring slices; or (5) Nyquist ghosts. Only the 20 components with the highest contributions were tested for rejection. 8.6 ± 2.7 components were rejected per run, representing 22.3% of total components (881/3954). [...] As noted above, one participant was excluded from fMRI data analysis due to excess motion (peak instantaneous translation > 2mm/volume).FSL analysis of fcMRI data was performed following and . Regions of interest (ROIs) in sensorimotor hand area were determined for each participant, by inverse-warping normative ROIs from standard space to each individual participant (). These normative ROIs were defined using data from a functional localizer task in which an independent sample of 17 healthy adults (mean age = 52 years, range 29–62, 4 female, 1 left-handed) performed an aurally-paced, thumb-finger sequencing task with eyes closed (). Spherical ROIs (5 mm radius) were centered on the peak activations in the left hemisphere (X = −38, Y = −24, Z = 54) and right hemisphere (X = 38, Y = −24, Z = 54).At the first level (i.e. for each functional scan), FSL's Featquery was used to identify the time course of changes in image intensity across all voxels in each ROI, after ICA and other preprocessing steps. These ROI time courses served as the explanatory (predictor) variables (EVs). Temporal derivatives were included to account for minor differences in timing. First-level contrasts of parameter estimates (COPEs) were calculated separately for each seed region, so that e.g. measurements of FC with the left-seed ROI were independent of any FC changes involving the right-seed ROI.The first-level analysis also included ten covariates of no interest to regress out non-neural changes, selected independently in each functional scan: (1) the time course of a voxel in white matter: anterior corpus callosum, or posterior corpus callosum if no voxel could be definitively identified as wholly in anterior corpus callosum; (2) the time course of a voxel in CSF: posterior horn of left lateral ventricle, or anterior horn if no voxel could be definitively identified as wholly in posterior horn; (3) the time course of a voxel outside the brain, superior to the anterior end of the third ventricle; (4) the time course of the whole-brain average; (5–10) motion covariates from MCFLIRT motion correction, encompassing three dimensions of translation and three of rotation. All covariates received the same temporal filtering as the fMRI data ().The top-level analysis followed previously successful methods for detecting FC changes across learning (). Two regressors were used. First, a participant regressor, to regress out consistent participant-specific effects; this regressor was constant for each participant, across all six runs (three runs × 2 sessions: pre- and post-training). The second regressor, which is the regressor of interest, modeled each participant's behavioral change across training. This applied regressor of interest, after orthogonalization with respect to the other regressor, was set to −(behavior)/2 for the scans on session 1, and +(behavior)/2 for the scans on session 2. Thus, the contrast of interest in the generalized linear model (GLM) was a graded variable based on each subject's behavior.For each seed region, three separate analyses were performed to identify areas that exhibited changes in functional connectivity: (1) between pre-training and post-training scans, with no explicit behavioral regressor; (2) between pre-training and post-training, and correlated with smoothness learning; (3) between pre-training and post-training, and correlated with smoothness retention. Speed learning and retention were omitted because of the absence of hand-specific speed learning (see Results 3.2); this choice was made before the start of MRI analyses. “Smoothness learning” for fcMRI analysis was defined as the ZNDH for each participant’s best session, mean-removed across participants. “Smoothness retention” was defined as ZNDH performance during NDH+6M session (i.e., performance difference between pre-training and 6-month follow up).At the top level, Z-statistic images were thresholded at Z > 2.3, with clusterwise correction at p < .05. Multi-fiducial mapping in Caret v5.64 (http://www.nitrc.org/projects/caret/) was used to overlay group statistical maps onto a population-average, landmark- and surface-based (PALS) atlas for visualization (). Slice views were produced using Mricron version 12/2009 (http://www.mccauslandcenter.sc.edu/mricro/mricron/).To measure the influence of hand-nonspecific fcMRI changes across the training period, the full analysis was repeated using seed regions defined by repetitive movements of the feet (instead of the hands). These ROIs were 5mm radius spheres centered on the voxel of peak activation in the left hemisphere (X = −4, Y = −28, Z = 72) or right hemisphere (X = 6, Y = −24, Z = 74) during toe flexion movements.Correlations between the two seed regions were measured by calculating the Pearson r between the average timecourse of all voxels in the left hemisphere hand seed and the average timecourse of all voxels in right hemisphere hand seed. The resulting r value was Fisher's z-transformed to stabilize variance and allow further statistical analysis.To determine whether training-related changes in FC were associated with reduced noise or increased signal, we calculated the temporal signal-to-noise ratio (tSNR; ) pre- and post-training. For each voxel, the tSNR was calculated as the mean signal value divided by the standard deviation of the voxel's signal intensity over time. A run's tSNR value was calculated as the average tSNR within a mask defined by all areas significantly correlated with the seed region at the group level, separately for each seed region. […]

Pipeline specifications