Overdiagnosis can be a problem. On the flip side, I wonder if adding the time dimension to the data (i.e., you could realistically have scans from every few weeks over the course of years) could significantly change that.
Instead of looking at a single snapshot of a person, you're now looking at trends over time. We probably don't have the analytical tools to effectively evaluate medical imaging with that time dimension at such scale (because I assume it would be rare for someone to get MRIs so frequently), but maybe with more data and study, we'll be able to more definitively distinguish benign quirks from real concerns.
Rather than a human comparing a couple of scans five years apart, you're talking about computationally identifying outlying regions in the data (a motion picture of the entire body) that are trending towards areas of concern.
I agree, hopefully it becomes just like any other cheap and easy measurement like blood pressure. Most of the time, small changes and variations will be determined to be normal but having a record and a measurement allows you to determine what is abnormal or a symptom of a disease. It's the difference between having a measurement and having nothing at all (if you never got a scan with CT or MRI).
This seems like something MLB teams would install in their locker rooms if it works.