Hi Fabian,
It seems that the question refers to the VBM template and to the group ICA. Do you know what is causing subjects to miss follow-up visits?
If subjects don't return due to reasons entirely independent from the study (e.g., weather, traffic, strikes, etc), then you can safely use for the template all three observations of the subjects that have complete data, i.e., those without any missing visit.
If subjects miss follow-up visits due to some other cause that may be associated with variables in the model, then the above will produce a biased template. You could consider taking just the first timepoint for all subjects. It will bias towards that timepoint, potentially penalising later timepoints, but shouldn't be too bad if analyses will compare slopes along time among different groups.
All the best,
Anderson