With all due respect in return, dementia is not like cancer. If you have cancer then, by de facto, there is a growth that shows on an MRI and the trick is spotting it. With dementia the atrophy produced that shows on an MRI is a secondary affect. The primary affect of the disruption in neural cell connections and the build up of abnormal proteins cannot be seen. Because of this the results of the MRI do not equate entirely to the start and/or progression of the disease. Sometimes there is a lot of damage seen on the scan, but hardly any symptoms and at other times there is considerable disruption to thought patterns and brain function, yet very little is seen on a scan. Some rare forms of dementia do not show anything at all on an MRI and have to have specialist forms of scans to reveal the problem. Also required are neuropsychology testing (there are lots of different tests in this category) that can be given, sometimes a lumber puncture and always the history given by people close to the person being tested. So, you see a scan is not the be all and end all of dementia, but simply one test in the midst of an arsenal of testing. Certainly AI can probably eventually be used for scans, but it does not lend itself to this analysis and I doubt it will replace everything else.
BTW, I am intrigued by your interest in AI - all your posts have been about this subject. Do you come from an AI background?