Quote (Neptunus @ May 3 2019 08:41am)

What happens if you commit malpractice, though? Who's liable?



IR seems to be the future. Endovascular and minimally-invasive procedures are increasingly gaining footage. I wonder how much AI will be taking over in medicine, though.

Quote (Bubbler @ May 3 2019 06:09pm)

The idea is that AI won’t “take”’over medicine but supplement it. At least that’s the way radiology is trying to portray it. And it does make some sense - AI has been reading ekg’s for years but no one trusts it without a clinician to interpret it, some doc’s won’t even read em until they’ve been verified by a cardiologist, imaging and radiology is much the same except in a few specialties like neurology where the scan has more value. IR and minimally invasive stuff is definitely going to have a continued bigger role in the future but surgeons are learning to do minimally invasive techniques. However surgery will never go away - after a certain point you need to physically see what is going on to make a diagnosis and fix the problem

Quote (Neptunus @ May 3 2019 11:43am)

Maybe not in the next few years, but if we were able to demonstrate the superiority of AI in diagnostics through studies, then i don't see why we should still trust a human over a machine. I believe it's mostly a medicolegal liability issue at that point. We already have plenty of diagnostic algorithms and once we have an algorithm we can program it, it's not technically impossible. I think the medical community is a bit divided with respect to AI. Some view it as an opportunity, some as a threat. It really depends on who you ask.

Procedural specialties aren't gonna be replaced anytime soon, but then again even in neurosurgery we have computer assisted procedures where a monitor shows what tissue to excise and what to spare and it's not that much of a stretch to imagine a machine performing the procedure. Not in the next few years but perhaps next few decades? I do agree that AI augmentation will be the first milestone and we will need humans to ensure everything goes as expected, but i don't believe it's too far-fetched to think it's replaceable possibly sooner than we expect. But enough AI rambling :D

Quote (Bubbler @ May 3 2019 06:50pm)

As I decided on radiology towards the end of my third year I obviously began to pay more attention to the AI in medicine argument, bc diagnostic radiology is obviously a target. I’m subscribed to various radiology outlets that constantly highlight landmarks like “mammography more accurate than radiologists in interpreting lesions” - the problem is these lesions May be in a 75 yr old and not actually become malignant for 20 years - so there needs to be clinical interpretation, EMR infrastructure needs to be overhauled to support AI, AI needs clinical accountability, medicine is slow to adapt technological change compared to other sectors as well

I’m not saying that what you are implying won’t happen, but I’ve read (a little) about the AI in imaging (at least) from the compsci side and extensively from my side, and I don’t think like it will happen for at least multiple decades, but I could very well be wrong and it could happen much faster. I doubt it though just based on the snails pace that medicine uses to adapt to change. But I’m obviously barely into the field, so I barely know anything

The big avenue that I see it making strides initially is by helping radiologists better stratify studies so more critical ones are interpreted first

Quote (Neptunus @ May 3 2019 12:31pm)

I'm not very familiar with pattern recognition AI personally, so i don't have much to say there. I've only seen some division in how quickly doctors believe that AI pattern recognition will become more reliable than that of humans. I know from first hand experience the ECG machine's own interpretations are indeed occasionally erroneus.

At the moment i'm doing internist and emergency stuff and more versed in emergency diagnostics. The things i'm talking about are like diagnosing a pneumonia or acute pancreatitis (the latter doesn't even require imaging if certain criteri are met). There are certain clinical markers that would be easily fed into a diagnostic program which just runs the exact same algorithms that humans go through with. Agreed on how slowly medicine adapts to change and it's partly caused by legal issues and partly by the fact that medicine is not engineering and something working on paper doesn't necessarily corremate with clinical efficacy. Indeed as you say we also need clinical interpretation that is based on evidence, and just gathering and processing the evidence required to be able to support automatization of diagnostics is a enough time-consuming.

Btw what does EMR stand for? The stratification idea sounds very plausible. I didn't think of it earlier, good point! Interesting stuff :)

Quote (Neptunus @ May 3 2019 01:31pm)

I'm not very familiar with pattern recognition AI personally, so i don't have much to say there. I've only seen some division in how quickly doctors believe that AI pattern recognition will become more reliable than that of humans. I know from first hand experience the ECG machine's own interpretations are indeed occasionally erroneus.

At the moment i'm doing internist and emergency stuff and more versed in emergency diagnostics. The things i'm talking about are like diagnosing a pneumonia or acute pancreatitis (the latter doesn't even require imaging if certain criteri are met). There are certain clinical markers that would be easily fed into a diagnostic program which just runs the exact same algorithms that humans go through with. Agreed on how slowly medicine adapts to change and it's partly caused by legal issues and partly by the fact that medicine is not engineering and something working on paper doesn't necessarily corremate with clinical efficacy. Indeed as you say we also need clinical interpretation that is based on evidence, and just gathering and processing the evidence required to be able to support automatization of diagnostics is a enough time-consuming.

Btw what does EMR stand for? The stratification idea sounds very plausible. I didn't think of it earlier, good point! Interesting stuff :)

Quote (Bubbler @ May 3 2019 07:40pm)

Electronic medical record! The 2 out of 3 methodology for pancreatitis makes perfect sense for an algorithm (I believe it’s a lipase, pain, and a ct where 2 of the 3 have to be met, iirc)? This kind of stuff I’ve felt like would be ripe for AI algorithms to take over (albeit I know less about it). The little that I’ve read about it taking over those sorts of tasks is the same thing that is bottlenecking it’s takeover in a lot of medicine - the clinical interpretation aspect of it. and in the case of a lot of these things (but not in radiology which is why I believe it is a prime target for AI) is the effect of bedside manner on patient satisfaction with their care.

Quote (Balla @ May 4 2019 02:43am)

Here's a good recentish paper on AI use heretofore
http://sci-hub.tw/https://doi.org/10.1038/s41591-018-0300-7
Obviously pathology & radiology will have the biggest implications sooner than others
My personal interest lies in BMIs though, and the machine learning that will take place for that. Massive implications for my field, one of my major interests.

Quote (ozzyarmy3 @ May 3 2019 04:05am)

I’ll stick to crna and not pay 100,000 a year in malpractice insurance.


😐

Quote (dark-soul @ May 7 2019 11:33pm)

Are you in a crna program?