I notice that the SPM maps from voxel-to-voxel results explorer and from ICA spatial summary are identical.  After back-projecting the ICA maps by adding as new ROI, the resulting seed-to-voxel SPM maps are very different for the same between-subject/between-condition contrasts. 

It is still unclear -

1) what is happening during preprocessing to the ICA maps that accounts for these differences
2) which approach is valid for assessing brain-behavior relationships?

Many thanks in advance,
Patrick