Isn't this pretty much what we expect from AI? Just like how it's learned to beat people at Chess and Go and Starcraft, it's going to be better than humans very soon when it comes to discrete problems: Is there a head bleed? How many lung nodules are there?
1) Humans are still going to need to over-read the AI for a long time to come purely for liability, even for those more mundane things.
2) In the study, AI was still consistently worse than the best radiologists... that will eventually change of course though.
3) I hope I don't offend breast radiologists by positing that AI is likely better suited for interpreting DM's than it would be for, say, body CT/MRI's - just my impression as a naive M2.
4) As technology eventually gets to the point where AI can do a reasonable job of interpreting complicated cross-sectional studies, radiologists will still be important for interpreting the AI's read. Just imagine an AI spitting out every abnormality seen on a body CT - it would be completely overwhelming and ridiculous. Clinicians will need help figuring out which findings are the most relevant and how to apply that information to patient care.
I think the role of radiologists is going to change a lot over the next 30 years... but I think they'll still be relevant and needed - just the role might entail less time spent counting lung nodules and more time spent being an AI shepherd / using critical thinking to work through fringe cases that stump the AI. The amount of data available and in need of interpretation is only going to continue increasing with new frontiers like radiomics opening up. Since radiology as a field is embracing AI and staying on top of it, it is more likely radiologists will be at the forefront of these changes and able to benefit from them.
I think the people who should really be shaking in their boots are the non-procedural IM/FM-type fields where AI-assisted midlevels could take over a lot of the bread-and-butter stuff.
Those are my musings as an M2 considering radiology...
@Mo991 I would love to hear your thoughts on these points, as someone with much better perspective on the field than I have. Am I off the mark?