Diagram showing the computation of the cross-eigenimages. Since the scores were computed with a distance-induced kernels, the cross-modality distance-correlation (DC) between the scores in one modality and the images of the other modality can be used to obtain the cross-eigenimages.
Similar to the case of the intra-modality spatial loadings, the individual scores in one modality can be distance-correlated to each of the voxels in the other modality to produce a set of maps, called "cross-eigenimages".
This procedure resembles a voxelwise linear regression and expresses canonical distributed-to-distributed configurations or views of the full whole-brain cross-distance-correlation between the two modalities. The strongest values in the cross-eigenimages can be interpreted as those regions in one modality that are maximally distance-correlated or nonlinearly associated with a spatially distributed set of regions in the other modality.