Figure 1

Illustration of the processing steps from raw data to the final brain map result. After pre-processing (bias correction, segmentation, normalization, modulation and smoothing), confidence intervals (CI) were obtained for each voxel of the brain. This allowed to construct a 'typical brain', i.e. a 5D matrix with 3 spatial dimensions, 1 dimension for the low and high bounds of the CI and the last dimension for number of CI). In this study, 5 CI were built after 3000, 3500, 4000, 4500 and 5000 bootstrap resamples. The 'typical brain' was therefore of dimensions 91*109*91*2*5 (91 voxels in x, 109 in y, 91 in z, 2 for upper/lower bound/5 for the 5 CI). Each voxel of each dyslexic subject was then classified as within or outside the 5 CIs. Results were then averaged, resulting in a percentage map of difference (PMD). A random attribution of scans to the control and dyslexic groups was used to compute the probability to find the observed values under H0, the null hypothesis according to which dyslexics and controls were sampled by chance from the same population.