Regarding the AAL template (which is most likely what you have already seen in the WFU pickatlas), I too have noticed that the AAL hippocampus definition tries to only cover grey matter. This exclusion leads to gaps and holes in the mask that I think you are referring to.
Given the anatomical variability between subjects in the hippocampus, it appears to me that the AAL mask attempts to be far too precise; moreover, it is not left-right symmetric. In the past, I have used AFNI's 'draw dataset' function to correct these problems with the AAL mask and used the resulting modified mask to define ROIs.
Alternatively, I've seen probabilistic anatomical masks that I think were used in FSL, and that may be worth checking out.
G Elliott Wimmer
Dept. of Psychology