STATISTICAL ANALYSIS OF REGIONS OF INTEREST
Percentage signal change and activation volume were the dependent variables used in the ROI analysis. To acquire percent signal change data, we first calculated the mean activation in each voxel for both rest and force blocks across each functional time series. The mean percent signal change for each voxel was calculated as the difference between the mean signal during the force blocks and the mean signal during the rest blocks divided by the mean signal during the rest blocks and then multiplied by 100. Therefore the output data at this level of analysis represented the percent signal change in each voxel. Resultant data sets were thresholded to include only voxels with a positive percent signal change located within a region of interest. In addition, we separately thresholded the percent signal change maps to include only negative voxels, and we performed the ROI analysis separately on these negative percent signal change voxels (deactivations).
The ROI analysis examined percent signal change and activation volume in the bilateral basal ganglia, the bilateral cortical areas, and the bilateral thalamus. These ROIs were drawn on a single Talairach-transformed anatomical image to form a template mask, and this template was overlaid on each subject’s Talairach-transformed functional image. The cortical ROIs included M1/S1, SMA, pre-SMA, and both ventral and dorsal premotor cortices (PMv and PMd, respetively) and these were based on the Human Motor Area Template (HMAT) created from a comprehensive meta-analysis of motor areas (Mayka et al. 2006
). The mask is publicly available (http://mcl.mvsc.uic.edu
In the basal ganglia, the regions included the caudate, anterior putamen, posterior putamen, external and internal portions of the globus pallidus (GPe and GPi, respectively), and subthalamic nucleus (STN). These ROIs are the same as those used in previous work (Vaillancourt et al. 2004
). Anatomical guidelines from previously published literature were used to help identify each basal ganglia nucleus (Yelnik 2002
). The centroid coordinate in Talairach space for the left-side ROIs are listed for each region (3dclust in AFNI).
(x = −11, y = 9, z = 11) is a curved structure with the rostral head being more voluminous than the body rostrally (Yelnik 2002
). It can be identified up to the level of the top of the ventricles. The medial border of the caudate nucleus is defined by the frontal horn or body of the lateral ventricle and the lateral edge by the anterior limb of the internal capsule (Ifthikharuddin et al. 2000
(x = −24, y = 2, z = 4) is limited medially on inferior sections by the globus pallidus and on more superior levels by the internal capsule (Ifthikharuddin et al. 2000
). Anteriorly, the anterior limb of the internal capsule separates the putamen from the caudate. Laterally, it is limited by the external capsule. The anterior and posterior parts of the putamen were differentiated on a slice-by-slice basis using the anterior border of the thalamus and the posterior border of the caudate as the dividing line.
is limited medially by the posterior limb of the internal capsule and laterally by the putamen (Ifthikharuddin et al. 2000
). It is divided into the globus pallidus internal portion (GPi) (x = −16, y = −4, z = 2) and the globus pallidus external portion (GPe) (x = −20, y = −4, z = 4). The GPe lies lateral to the GPi and is almost twice as large (Yelnik 2002
(x = −11, y = −14, z = −3) lies ventral to the thalamus, medial to the peduncular portion of the internal capsule, and lateral and caudal to the hypothalamus. It is lateral to the red nucleus and dorsolateral to the substantia nigra in the coronal plane (or anteromedial in the axial plane) (Dormont et al. 2004
). The size of the STN may be smaller than reported in the Talairach and Tournoux atlas, particularly in the medial–lateral direction (Richter et al. 2004
The thalamic ROIs included the ventrolateral (VL), ventroanterior (VA), lateral ventroposterior (VPl), medial ventroposterior (VPm), and centromedial (CM) nuclei. Descriptions of each of these ROIs follow.
(x = −10, y = −6, z = 8) is the most anterior component of the ventral thalamic nuclei, lying rostral to the VL nucleus. It is lateral to the AV nucleus in the superior portion of the thalamus in the axial plane, but it extends medially to the internal capsule in the inferior portion of the thalamus (Wiegell et al. 2003
VL nucleus (x = −12, y = −12, z = 10) lies caudal to the VA nucleus in the axial plane on the ventral edge of the thalamus, medial to the internal capsule. It is lateral to the dorsomedial (DM) nucleus and extends from the superior portion of the thalamus to the inferior boundary. The VL nucleus is bordered rostrally by the VA nucleus and caudally by the VPl nucleus of the thalamus.
VPl nucleus (x = −18, y = −19, z = 5) and VPm nucleus (x = −14, y = −19, z = 5). The ventral posterior nucleus of the thalamus is split into lateral and medial portions. The lateral portion of the ventral posterior nucleus is medial to the internal capsule and lateral to the ventral medial posterior nucleus. Both nuclei lay inferior to the lateral posterior nucleus and extend to the interior edge of the thalamus.
CM nucleus (x = −8, y = −19, z = 3) is bordered medially by the caudal edge of the DM nucleus and laterally by the VPm nucleus of the thalamus. The CM nucleus is also inferior to the superior portion of the DM nucleus and the VPm nucleus and extends to the inferior edge of the thalamus. The caudal edge of the CM nucleus borders the posterior nucleus of the thalamus.
Next, we quantified the average percent signal change and activation volume within the aforementioned ROIs. The average percent signal change was calculated across all positive (or negative) percent signal change voxels within an ROI. Percentage signal change of each individual ROI was calculated at 5, 20, 40, 60, or 80% MVC for each individual subject, separately for activations and deactivations. In calculating the activation volume, we first used a general linear model (3Ddeconvolve, AFNI) to regress the voxel time-series data to a simulated hemodynamic response function for the force and rest blocks included in each of the five MVC force level scans. This level of analysis resulted in a regression coefficient for each voxel with an associated P value. The activation volume was calculated by counting the number of significantly activated voxels that had a t-value >3 (P < 0.005).
The mean percent signal change data and activation volume for individual subjects were analyzed using separate repeated-measures ANOVAs for each ROI. The within-subjects repeated factors were hemisphere (left or right) and target force level (5, 20, 40, 60, and 80%). Separate one-way ANOVAs were performed for each hemisphere of the regions that showed a significant force × hemisphere interaction. A paired t-test was performed at each force level to determine the locus of the interaction between each hemisphere of GPe. All statistical tests were evaluated as significant if the P value was <0.05.
The positive percent signal change data in each ROI were further analyzed using a correlation analysis between each ROI and the three kinetic parameters under investigation (force, rate of change of force, duration of force). All correlation analyses were computed using data from the left hemisphere only. The group mean percent signal change data at each force level provided five data points for comparison within each ROI. The Pearson product moment coefficient of correlation (r) was calculated between the group mean percent signal change in each ROI and each kinetic parameter (Statistica, 6.1). A correlation was considered statistically significant when P < 0.05.
VOXELWISE GROUP ANALYSIS FOR ROI PRESENTATION
We constructed group maps of GPi, STN, SMA, M1/S1, and thalamus for visualization of activation within the ROIs. We first normalized the signal in each scan of each subject by dividing the instantaneous signal in each voxel at each point in the time series by the mean signal in that voxel for each subject’s scan. This process was repeated for each scan for each subject. Therefore at the end of this process, the blood-oxygen-level– dependent (BOLD) signal in each functional data set for each subject was normalized around a baseline value of 100. After this, we applied a Gaussian filter to the resultant data sets [full-width-half-maximum (FWHM) at 5 mm]. Then, we used a general linear model (3Ddeconvolve, AFNI) to regress the time-series data to a simulated hemodynamic response function for the force and rest blocks included in each of the five MVC force level scans. We then performed a group analysis of each force level scan. In the group analysis, we first used a mixed-effect two-way ANOVA with the respective MVC force level as the fixed factor and the subject as a random factor. All group maps were then masked using the ROIs generated a priori, to present only those voxels that were contained in the ROIs that we used for our hypothesis testing. The maps were thresholded so that all included voxels had a t
-value >3 (P
< 0.005, uncorrected). The uncorrected statistical threshold was used for the group analysis because we had specific hypotheses regarding the basal ganglia nuclei. The analysis using uncorrected P
values can be more informative when studying small brain targets that produce weak BOLD signals like the basal ganglia regions (Turner et al. 2003
; Vaillancourt et al. 2004
; Wu and Hallett 2005
). The thresholded masks were coregistered to an individual anatomical map in Talairach space.
To construct the three-dimensional (3D) functional surface maps of the cortex, the data sets were corrected for type I error by Monte Carlo simulation (AFNI alphasim, http://afni.nimh.nih.gov/afni/doc/manual/AlphaSim/
). The cortical activation maps were thresholded at t
= 3 with an activation cluster minimum of 500 μ
l to give a significantly low probability for type I error (P
< 0.01, corrected). The data were then masked to display only those voxels within M1/S1 or SMA. The corrected functional maps were coregistered to a Talairach version of the N27 brain data set (MNI, http://www.bic.mni.mcgill.ca
; UCLA, http://www.loni.ucla.edu
). The surface maps displaying cortical data were created using the AFNI surface mapper, SUMA (http://afni.nimh.nih.gov/afni