These systems are basically banned from working as actual replacements for radiologists, so it's no surprise they aren't yet. We have repeatedly proved the supe

These systems are basically banned from working as actual replacements for radiologists, so it's no surprise they aren't yet. We have repeatedly proved the superiority of expert systems (in the 90s) and AI and select medical tasks. However, there is a legal monopoly (in the US) that requires most medical tasks to be performed by expensive doctors. If people were able to use these tools directly, we could see dramatically better results because we would be giving people decent healthcare at basically zero cost. Cost is by far the biggest problem in healthcare today. Low cost would change behavior in a large number of medical tasks, and early detection of cancer is the most obvious. If you could get a mediocre readout for free, you would probably do so more often. Cancer in particular is almost entirely an early detection problem. Using AI to assist radiologists is probably never going to be a huge thing. Just like AI assisted truck driving is never going to be huge (because it doesn't solve the core problem).