scanner effects across vendor could be larger than disease effects. You can attempt an analysis but you will have a tough time to publish it given these differences and no real way to control for scanner effects. Maybe scan 5-10 people on all these scanners and show that scanner effects remain small for the regions of interest (of course this would not help i there is an interaction, e.g. with diesease or age) but at least it could help convince reviewers. Also if you only analyse the first 2 time points  and find large differences, you could argue that they are real, because the same scanner should not course large effects. But who knows (different head coils? Different callibrations, software versions?…). Better to test with a few subjects. 

Best, Martin