2026-09-29 11:48 UTC
Replies (3)
-
@markmetz@sfba.social 2026-09-29 11:52
“Don't even get me started about how expensive it would be from a compute and capital perspective to train new generations of models on cross-sectional images with >GB files sizes, cross reference those models with the underlying clinical data, gain regulatory approval for mass access to imaging data, and the inevitability of model drift as new generations of imaging hardware and procedures continue to roll out. Oh, and legal liability, which would require a congressional act of god to immunize them from. Trust me: I will be doing this field for a long time to come, and you too will be doing yours. Do not believe these out of touch plutocrats.” #NoAI
-
@markmetz@sfba.social 2026-09-29 11:50
“I'm truly just blown away by how little impact AI has had in our field, even when it comes to pure diagnostics. My hospital employs a limited number of "models" for basic tasks like pulmonary nodule detection, and even then they suck to such a point that I stopped using them early on in my residency. Beyond that, I have never heard of anything remotely approaching the point of analyzing CT, MR, US, PET, or catheter angiography accurately and producing a complete report. The technology is not just in "early stages"...it's plainly nonexistent. And the "cutting edge" research that I review ain't exactly making much progress (it's usually for much more banal bullshit, like fixing reporting errors, and triaging critical findings...which is all well and good, but only modestly impactful in the setting of a true clinical workflow).” more->
-
@Julianoe@mastodon.xyz 2026-09-29 12:16
@markmetz@sfba.social Interesting. When it comes to "AI talking jobs" I don't think the majority of people think of radiologists though. A lot of dequalified jobs today are simply asking of humans to be button-pushers. Those are the ones that are going to vanish. But it's not solely because of ai. The dequalification process has started many years ago. It's simply the new step.