Sources estimated with dSPM cannot be directly used to reconstruct the MEG/EEG signals, because the dSPM sources are the minimum-norm sources normalized to the noise in each source location. Thus at each source location the sources have been scaled in a different way, which besides removing the physical units, changes the magnitude relationship among sources, so it is not appropriated to use the forward model on these normalized sources.
If you want to reconstruct the MEG/EEG signals, you could use the source estimation using MN imaging and the current density map option.
@Sylvain, any thoughts on this?
@Jixiang, I just realized this is a repost of the question, the previous which as not answered is now delete. Apologies for the answer delay.