Diagram showing the computation of the intra-modality spatial loadings, which express the spatial information encoded in the Canonical Distance Correlation (CDC) scores. Since the scores are computed with a distance-induced kernels, the intra-modality distance-correlation between the scores and the images can be used to obtain the spatial loadings.
The eigenimages or spatial loadings can be computed by performing a voxelwise distance-correlation between each voxel in the image dataset and the corresponding CDC scores, a procedure that resembles the typical intra-modality linear projection step used in Singular Value Decomposition and Canonical Correlation Analysis.
Taken together, the ordered pairs of eigenimages produce a spatial representation of the CDC scores in the original image space. The high values in the eigenimages can be interpreted as those regions that show the strongest contribution to the scores.