Fig. 3.

Overview of the leave-one-out cross-validation regression approach. (a) For each participant we estimated regression coefficients (‘beta’) describing how functional activity related to structural connectivity with 148 cortical targets. This was implemented for each striatal subregion and functional contrast independently. (b) The regression coefficients were averaged across the n-1 participants (‘betaCV’ indicating cross-validated beta) and multiplied by the n-th subject's connectivity matrix to predict the contrast coefficient in each voxel. The Pearson correlation between predicted and observed coefficients was recorded and the approach repeated for each participant, yielding n correlation coefficients for each contrast and striatal subregion. A Pearson's r significantly greater than zero indicates differences in functional responses between voxels are predicted from differences in structural connectivity.