I think that "3dTcorrelate" provides the most similiar function for this purpose. However, for canonical correlation, the calculation should not be between corresponding voxels but between corresponding ROIs of "xset" and "yset". Namely, I want to move a spherical ROI (user defined radius) in the brain with the output reporting voxel-wise canonical correlations. I was trying to do this with matlab but it's very slow. I am wondering if "3dTcorrelate" can have a new option of "-canonical" in future updates.