Skip to main content
. 2020 Apr 22;41(12):3235–3252. doi: 10.1002/hbm.25011

FIGURE 2.

FIGURE 2

A layout of the inference scheme. For a subset of n subjects, an explanation map was computed, representing the contribution of each voxel to the model's output. Each saliency map was first registered to the subject anatomical image, then it was transformed to the MNI space. Next, each volume was smoothed with a 3D Gaussian. Finally, all the volumes were averaged to create a population‐based explanation map