119 CEREBRAL ADAPTATION ASSOCIATED WITH PERIPHERAL NERVE RECOVERY IN NA 6 Analyses Preprocessing of task-fMRI data MRI data were preprocessed using FSL 5.0.11 (FMRIB’s Software Library, http://www. fmrib.ox.ac.uk/fsl).129 We first removed non-brain structures from the structural image, 130 realigned the functional images, 131 applied smoothing (Full Width at Half Maximum (FWHM) = 3 mm) and grand mean scaling. Next, we registered the functional images to the structural image and standard MNI152 space132-135 and removed motion-related noise with the FSL tool ICA-AROMA. 136 After manual inspection and appropriate reclassification of the components generated by ICA-AROMA (n=100), we applied non-aggressive denoising. 137 We performed nuisance regression on the denoised images, including 24 motion parameters, 138 and cerebrospinal fluid and white matter regressors. Finally, we applied a temporal high pass filter of 0.01 Hz and additional smoothing with a 5.2mm FWHM Gaussian kernel (total smoothing of 6mm FWHM). See Lustenhouwer et al. 2 for detailed description of preprocessing. First level task-fMRI analyses We performed further analyses with FSL version 6.0.1. 129 We modelled blood-oxygenlevel dependent (BOLD) activation on the first level (subject and session specific) with a general linear model (GLM) that included factors BIOMECHANICAL COMPLEXITY, POSTURAL CONGRUENCY and LATERALITY. 131 All regressors of interest included correct responses only. Events were set at stimulus onset and the duration was the subject- and session-specific median RT across trials. We additionally included three regressors of noninterest: limb-repositioning between blocks (7000ms duration), incorrect, and missed trials (duration of median RT). BOLD activation was modelled by convolving all regressors with a canonical hemodynamic response function and its temporal derivative. 142 We performed pre-whitening followed by voxel-wise fitting of the GLM. 131 The resulting contrast of parameter estimate (COPE) images were registered to MNI152 standard space. 132-135 Main hypothesis testing We tested for possible group differences in the change in cerebral and behavioral responses from baseline to follow-up for the affected (right) vs. the unaffected (left) limb. We focused on a priori defined ROIs124 in the right extrastriate cortex and bilateral parietooccipital sulcus during biomechanically complex trials, following our previous findings showing decreased activity in NA versus healthy controls in these regions. 2 Group level task-fMRI analyses To investigate the effect of factors TIME (baseline, follow-up), LATERALITY (right, left) and GROUP (usual care, multidisciplinary rehabilitation) on imagery-related activity, we first calculated the patient-specific difference in activity between timepoints (follow-up minus baseline COPE image) for the contrast of interest (i.e. right > left, complex trials only). We then performed non-parametric threshold-free cluster enhancement (TFCE) based permutation testing (5000 permutations), 143 correcting for multiple comparisons with a conservative FamilyWise Error of p < 0.05. The GLM in this analysis included factor GROUP, with the patient-specific follow-up minus baseline COPE images as input. We performed