Saw a demo of an AI that reads medical scans at a hospital in Boston

I was at Massachusetts General Hospital last month for a conference and they showed us an AI system that spots early signs of tumors in CT scans. The doctor running the demo said it caught a tiny lung nodule a human radiologist had missed just the week before. It made me think about how these tools could really speed up diagnoses. Has anyone else seen AI being used in medical imaging where you work?

4 comments

4 Comments

owens.anthony1mo ago

Wait, a radiologist actually missed it? That's wild to think about - I mean, @smith.parker, you hear stories but seeing it in person is something else. The doctor at the demo said the AI flagged a shadow the size of a sesame seed that the human eye just skipped over somehow. Makes you wonder how many tiny things have been slipping through the cracks, you know? They claimed the system catches things like that about 20 percent more often than a regular read too. Kinda changes how you look at false positives when the alternative is missing something real.

smith.parker2mo ago

What's the false positive rate like on that system? I've heard some of these AI tools flag so many harmless shadows that it creates more work for doctors, not less. Did they talk about how often it cries wolf?

jana_black2mo ago

Yeah the "cries wolf" thing is exactly what I worried about too, @smith.parker. But the speaker showed data where the latest version flags way fewer clear lungs as suspicious, like a single-digit percentage now. It's gotten better at telling the difference between a real shadow and just a weird angle.

ivan7742mo ago

Hold up, let's not get too excited about a single-digit false positive rate. That still means it's wrongly flagging, what, maybe 5 out of every 100 scans? In a big hospital, that's a ton of extra work. Doctors still have to check every single alarm, and now they're wasting time on clear scans. Plus, what if the one it misses is the real cancer because it got too good at ignoring shadows? Seems like we're trading one problem for another.