I got up at 4am to run scans in grad school because the even the research scanners were booked solid, never mind the clinical ones. Scanners are really expensive: the machine costs a lot, it needs fairly particular conditions (shielded room, cryogens), and trained operators and analysts.
There's maybe a market for fleecing the extremely wealthy into undergoing MRI fishing expeditions, but you'd have to charge a fortune if you're going to buy scanners to do it.
There are hospitals that have switched brands of scanners over protocol times.
In the case of blood chemistry that is caused by fluctuations due to time of day, time of month, season, what you recently ate and so one.
In the case of MRI the problem is that internal organs shift to the point that it is impossible to automatically compare. And a experienced doctor comparing the scan millimeter for millimeter takes hours and would be very expensive. Even just tracking the volume of a known tumor in the brain (that shifts very little) to track the effect of radiation is tedious, time consuming and error prone.
I didn't even know that is a thing, I just sorta assumed they stayed put.
Seems like a good area to apply machine learning if we had enough data?
There are whole studies done to try and quantify this, and it requires sending "human phantoms"- test subjects- around to different scanners, which requires you to physically take one subject around to several scanners operated by different people and stick them in there for a while. It's really involved.
In the same way astronomers can make new discoveries by digitizing photographic plates from a century ago.
Scanners cost a lot: ~4M for a 3T (the current 'standard') and much more for higher field strengths. We just got a 7T and, with all the trimmings (scanner, console, coils, the shielded room to hold it all, training, etc), it's in the $10-15M range.
As a result, time on these things is already in pretty high demand, even at facilities that are used only for research. The projects could be reprioritized, of course, but most of the on-going research is much less speculative than "someday, this can be training data for something."
It's not even clear to me that old MRI data will be particularly worthwhile. Photographic plates were, until recently, about as good as digital sensors. New MRIs are much, much better than old ones, and the tech will only improve more. Longitudinal data is useful in both fields, but for astronomy, there's no way to rewind time, so old plates are the only possible source for events from the past. We get new old people, however, every day!
Where is that? Might be worth the trip in some cases.
I know Yale also has a 7T, as does the NIH, Stanford, and MGH (I think--it's somewhere in Boston). The University of Minnesota has THREE which is crazy.
Wouldn't it be more useful as a baseline? Scan once and store it, then scan every few years and compare. If any of the "anomalies" has grown or new ones have appeared between scans, then they might be worth a second look?