Hi Alain,
I don't think it will be a huge problem as it appears. Assemble a design with 14 EVs, these being:
EV1: Scanner A, group 1
EV2: Scanner A, group 2
EV3: Scanner A, group 3
EV4: Scanner A, group 4
EV5: Scanner A, group 5
EV6: Scanner B, group 1
EV7: Scanner B, group 2
EV8: Scanner B, group 3
EV9: Scanner B, group 4
EV10: Scanner B, group 5
EV11: Scanner C, group 1
EV12: Scanner C, group 2
EV13: Scanner C, group 3
EV14: Scanner C, group 4
where "group 5" doesn't have any data in scanner C. Different scanners may also have different variances so you may as well define one exchangeability block (EB) and one variance group (VG) per scanner, and run this in PALM.
The contrasts for the between-group differences can be constructed separately for each of the 3 scanners, and then combined (except for group 5, which will not belong to an analysis encompassing all three scanners). Same applies for the interaction contrasts: are possible but not for the group 5 when scanner C is considered.
Hope this helps.
All the best,
Anderson