Fig. 2.

Model predictions correlate with observed activations across subjects. (a) Visualising the overlapping activation between two example subjects' maps (i.e., subjects (a) and (c) from Fig. 1). For illustrative purposes, these maps have been thresholded and binarised. Binary overlaps were quantified using the Dice coefficient, which is highest along the diagonal (representing each subject's observed and predicted maps). The off-diagonals pair different subjects' maps. (b) The pattern in (a) generalises across most subjects, producing a similarity matrix of correlated maps. The strong diagonal shows that similarity is strongest within subjects' maps. Pixel intensities show correlations (Pearson coefficients); these have been normalised across the rows and columns to make the results comparable between scans. (c) Histograms of the correlations in (b). The off-diagonals (blue) appear normally distributed around zero, the diagonals (yellow) are not. The diagonal and off-diagonal distributions differed significantly (t(102.8) = 25.09, p ≪ 0.001).