Rads to Anesthesia?

[DrfluffyMD]

1 Do you practice anesthesiology? There is a forum for radiology I assume.

2 Maybe because there is no anesthesiologist making 3M. Probably not even 1M doing OR.

1. I do not practice anesthesia, but that's not the topic of discussion here. we seemed to be discussing radiology, or what you thought radiology is like that is not a generalizable situation. Frankly, I am not sure if this is something that even exist for more than once in someone's life on a particularly bad call or outside of abusive teleradiology practice.

2. My statement about anesthesia is what I IMAGINED how outlandish your statement about radiology would sound to you if translated to anesthesia terms. Again, I am not an anesthesiologist, just trying to say the scenario you are speaking of (AI taking over radiology imminently, random telerad reading 1000 studies a day day in and day out as a living) are incredibly outlandish.

---

Members don't see this ad.

 

[urge]

1. I do not practice anesthesia, but that's not the topic of discussion here. we seemed to be discussing radiology, or what you thought radiology is like that is not a generalizable situation. Frankly, I am not sure if this is something that even exist for more than once in someone's life on a particularly bad call or outside of abusive teleradiology practice.

2. My statement about anesthesia is what I IMAGINED how outlandish your statement about radiology would sound to you if translated to anesthesia terms. Again, I am not an anesthesiologist, just trying to say the scenario you are speaking of (AI taking over radiology imminently, random telerad reading 1000 studies a day day in and day out as a living) are incredibly outlandish.

Let's just agree to disagree.

 

[AdmiralChz]

Interesting - NightHawk Service also offers after-hours and weekend TTE and even TEE reading. Anyone know of hospitals that use that service?

One HCA hospital in our metro area is actively negotiating with similar companies to provide 24-hour coverage which would push the radiologists completely out. Not popular, but it's all about the bottom line and it's been a nasty negotiation with the radiology group there. IR is covered by a separate group. Food for thought.

My friends in radiology residency are very concerned about the AI thing, they are actively involved in quality, cost analyses and metric improvements in the hospital to constantly be involved and have a seat at the table. The most successful anesthesia departments (academic and PP) do the same. It goes beyond what your actual j

 

[urge]

Interesting - NightHawk Service also offers after-hours and weekend TTE and even TEE reading. Anyone know of hospitals that use that service?

One HCA hospital in our metro area is actively negotiating with similar companies to provide 24-hour coverage which would push the radiologists completely out. Not popular, but it's all about the bottom line and it's been a nasty negotiation with the radiology group there. IR is covered by a separate group. Food for thought.

My friends in radiology residency are very concerned about the AI thing, they are actively involved in quality, cost analyses and metric improvements in the hospital to constantly be involved and have a seat at the table. The most successful anesthesia departments (academic and PP) do the same. It goes beyond what your actual j

Don't talk about "friends". It does not bode well here.

 

[pgg]

This gets thrown out there a bunch but to the best of my knowledge it's not true. I'm fairly certain you have to have a medical license in the US in order to get reimbursed for reads. Many places use nighthawk services but they are based in other parts of the US. Or like one radiology practice my wife knows, rents an apt in Paris and sends people over for a few weeks at a time to read overnight.

But it's not like overnight reads are being outsourced to Indian radiologists or something.

You need a drivers license to drive a car in the US. Today. That's not going to stop software-driven cars. Especially since the driverless cars will soon be safer than human drivers (if they're not already).

There will come a day when diagnostic radiology is owned by machines. The liability argument will be moot when either (a) the machines do it better than the people, and/or (b) it becomes more cost effective to pay a 1/2 a radiologist's salary to cover the software license fee and the liability insurance premium.

Insurance is just a numbers game, and you can bet that when it turns out that the software messes up less than the humans, the insurance industry will sell a profitable contract. It's what they do.

The only question is whether it'll happen during the career arc of today's radiologists-in-training. You can't really believe that diagnostic radiology is going to be done by humans in 2050, can you?

This. The day AI takes over radiology is the day that AI takes over many other specialties in medicine, as well as other industries. They cant even come up with an accurate EKG machine to replace the need for cardiology interpretation.

100% of my patients these days get a 12-lead EKG. 0% of those EKGs get read by cardiologists before the patients go to the OR. The machines do a pretty good job. It's rare that my read is different than the machine read, and when I do, it's typically something small.

I think the cardiologists look at them eventually, in what amounts to a spectacular waste of their time and healthcare resources. Fortunately, cardiologists have useful skills beyond reading EKGs. Their value is largely what they DO about the data that has been collected.

Machine learning will streamline and assist radiologists, but a complete takeover? Lol. And who is gonna have medical liability? Watson?

Watson's insurer. You know, the one paid with premiums that amount to less than the cost of hiring radiologists.

It's absolute folly to think that a system that costs less and does a better job than a human won't be embraced by the ones paying the bill. Of course it will.


Anesthesiology has its issues, and the near term job outlook for anesthesiologists may or may not be good, but there's always going to be a human doing it because of the human-to-human patient interaction, and close integration of realtime mechanical tasks, data gathering, data interpretation, and decision making - all fed back into the immediate loop of mechanical tasks that need to be done, pronto.

The field of diagnostic radiology, lacking the human connection or need for any associated movement of tools or stuff, is wholly different. The entirety of the specialty is interpreting digital images acquired by other people, correlating those findings with patient history acquired by other people, and reporting the findings and (perhaps) recommendations to be acted upon by other people.

You think a piece of software won't ever do that? We're not talking about flying cars, room-temperature fusion reactors, or strong AI.

Interesting - NightHawk Service also offers after-hours and weekend TTE and even TEE reading. Anyone know of hospitals that use that service?

Who acquires the TEE images to be read remotely?

 

[DrfluffyMD]

You can't really believe that diagnostic radiology is going to be done by humans in 2050, can you?

Fact 1: In 1989, the Buran orbiter flew to orbit and landed back, all controlled by AI.

Fact 2: Autopilot can land planes better than human pilots in certain conditions.

Fact 3: we still have human pilots.

Fact 4: Automation in anesthesia (unfortunately, and wrongly) have progressed beyond automation in radiology. There is already a FDA approved automated system.

This system is infact, so unpopular it got pulled off market. Not surprising as clinician prefer real people.

Fact 5: No machine or program to date have been able to replace a radiologist in the imaging chain.

Fact 6: To completely replace a human radiologist, a machine GENERAL intelligence that is self aware is needed, or a human + narrow AI will always beat out a narrow AI.

We still don't have flying cars, and I don't anticipate we'll have artificial general intelligence / machine intelligence / separate intelligent specie of machine being by 2050.

If we do, we have a bigger problem.

Meanwhile, you can convince your surgeon to accept report for machine.

I am going to IR and plan to do 100% of it. I sure as hell don't want a machine to replace my anesthesia colleagues, nor radiology colleagues.

 

[urge]

I am going to IR and plan to do 100% of it.

The only smart sentence in your whole post.

 

[by604]

Because you're considering switching to anesthesia; a field with a future even more bleak than radiology. Hey, both fields DO have something in common: you have to do multiple fellowships to be employable.

most fields do fellowships anyways these days soo...

 

[DrfluffyMD]

The only smart sentence in your whole post.

I see you resorted to personal attacks rather than rational debate. I thought we "agreed to disagree"?

Either way, perhaps I'll bring this ignorance up to the anesthesia colleague I work with and watch them laugh along with me.

 

[pgg]

You'll be fine in IR, because your hands will be needed to do things that your trained brain tells them to do.

You'd probably even be fine for a career in diagnostic radiology, because computer takeover of image interpretation isn't imminent.

But your litany of facts failed to address the most important part of my post:

The field of diagnostic radiology, lacking the human connection or need for any associated movement of tools or stuff, is wholly different. The entirety of the specialty is interpreting digital images acquired by other people, correlating those findings with patient history acquired by other people, and reporting the findings and (perhaps) recommendations to be acted upon by other people.

You think a piece of software won't ever do that?

The part of pathology that examines slides under a microscope is similarly vulnerable.

 

[vector2]

Fact 1: In 1989, the Buran orbiter flew to orbit and landed back, all controlled by AI.

Fact 2: Autopilot can land planes better than human pilots in certain conditions.

Fact 3: we still have human pilots.

Fact 4: Automation in anesthesia (unfortunately, and wrongly) have progressed beyond automation in radiology. There is already a FDA approved automated system.

This system is infact, so unpopular it got pulled off market. Not surprising as clinician prefer real people.

Fact 5: No machine or program to date have been able to replace a radiologist in the imaging chain.

Fact 6: To completely replace a human radiologist, a machine GENERAL intelligence that is self aware is needed, or a human + narrow AI will always beat out a narrow AI.

We still don't have flying cars, and I don't anticipate we'll have artificial general intelligence / machine intelligence / separate intelligent specie of machine being by 2050.

If we do, we have a bigger problem.

Meanwhile, you can convince your surgeon to accept report for machine.

I am going to IR and plan to do 100% of it. I sure as hell don't want a machine to replace my anesthesia colleagues, nor radiology colleagues.

Fluff, I tend to agree that ai rads is probably further away than we think, but why do you think strong ai or agi is needed for diagnostics? Given a large enough normal variant / pathology database, clinical history, and sufficiently advanced image processing, shouldn't a machine be able to do say 99.5% of body ct or neuro?

 

[Hoya11]

Fluff, I tend to agree that ai rads is probably further away than we think, but why do you think strong ai or agi is needed for diagnostics? Given a large enough normal variant / pathology database, clinical history, and sufficiently advanced image processing, shouldn't a machine be able to do say 99.5% of body ct or neuro?

Were not going to see a computer replace an anesthesiologist or a radiologist in our lifetimes. I do hope we see a computer replace a driver though. Arguing that either job is so easily automated is insulting.

 

[deleted547339]

You need a drivers license to drive a car in the US. Today. That's not going to stop software-driven cars. Especially since the driverless cars will soon be safer than human drivers (if they're not already).

There will come a day when diagnostic radiology is owned by machines. The liability argument will be moot when either (a) the machines do it better than the people, and/or (b) it becomes more cost effective to pay a 1/2 a radiologist's salary to cover the software license fee and the liability insurance premium.

Insurance is just a numbers game, and you can bet that when it turns out that the software messes up less than the humans, the insurance industry will sell a profitable contract. It's what they do.

The only question is whether it'll happen during the career arc of today's radiologists-in-training. You can't really believe that diagnostic radiology is going to be done by humans in 2050, can you?




100% of my patients these days get a 12-lead EKG. 0% of those EKGs get read by cardiologists before the patients go to the OR. The machines do a pretty good job. It's rare that my read is different than the machine read, and when I do, it's typically something small.

I think the cardiologists look at them eventually, in what amounts to a spectacular waste of their time and healthcare resources. Fortunately, cardiologists have useful skills beyond reading EKGs. Their value is largely what they DO about the data that has been collected.




Watson's insurer. You know, the one paid with premiums that amount to less than the cost of hiring radiologists.

It's absolute folly to think that a system that costs less and does a better job than a human won't be embraced by the ones paying the bill. Of course it will.


Anesthesiology has its issues, and the near term job outlook for anesthesiologists may or may not be good, but there's always going to be a human doing it because of the human-to-human patient interaction, and close integration of realtime mechanical tasks, data gathering, data interpretation, and decision making - all fed back into the immediate loop of mechanical tasks that need to be done, pronto.

The field of diagnostic radiology, lacking the human connection or need for any associated movement of tools or stuff, is wholly different. The entirety of the specialty is interpreting digital images acquired by other people, correlating those findings with patient history acquired by other people, and reporting the findings and (perhaps) recommendations to be acted upon by other people.

You think a piece of software won't ever do that? We're not talking about flying cars, room-temperature fusion reactors, or strong AI.




Who acquires the TEE images to be read remotely?

Your ekg argument is terrible. Sure, in the pre-op setting, it may be OK. In the ER, I would say that I disagree with at least probably 10%. I mean BIG things. Calling STEMIs that aren't there, missing STEMIs, missing heart blocks/arrhythmias.

 

[urge]

I see you resorted to personal attacks rather than rational debate. I thought we "agreed to disagree"?

Either way, perhaps I'll bring this ignorance up to the anesthesia colleague I work with and watch them laugh along with me.

Says the guy who keeps calling me ignorant and yet knows I'm right. Thus is running for safe groung, IR.

You are welcome.

20 years down the line when you still have a job you will thank me.

 

[DrfluffyMD]

Says the guy who keeps calling me ignorant and yet knows I'm right. Thus is running for safe groung, IR.

You are welcome.

20 years down the line when you still have a job you will thank me.

I am sorry, I am not really understanding what you are talking about? Please elaborate.

 

[urge]

I am sorry, I am not really understanding what you are talking about? Please elaborate.

I'll elaborate.

 

[kjamess]

100% of my patients these days get a 12-lead EKG. 0% of those EKGs get read by cardiologists before the patients go to the OR. The machines do a pretty good job. It's rare that my read is different than the machine read, and when I do, it's typically something small.

I think the cardiologists look at them eventually, in what amounts to a spectacular waste of their time and healthcare resources. Fortunately, cardiologists have useful skills beyond reading EKGs. Their value is largely what they DO about the data that has been collected.

your EKG example is irrelevant. Why? Because these preop EKGs in asymptomatic patients rarely change management, and aren't indicated in many instances to begin with. And guess what? These patients usually have already received preop clearance from a cardiologist in the first place! So no, I don't care that you are fully dependent on a meaningless automated read.

What is interesting, however, is how quickly things change post op in the PACU when the patient now develops some symptoms. Still feel comfortable depending on the "non-specific ST/T wave abnormalities" read from the machine? I have several experiences from residency where anesthesia would consult much too late for an evolving STEMI that happened hrs ago (based on the initial EKG) but no one acted upon because the automated read said "normal sinus rhythm. normal ecg". As I said before, sensitivity is in the 60s for automated read detection. Its not what it finds, its what it doesn't.

 

[DrfluffyMD]

I'll elaborate.

I figured you must be very early on in your training and cannot come up with a good discussion. Good luck in your future career through medicine!

 

[DrfluffyMD]

your EKG example is irrelevant. Why? Because these preop EKGs in asymptomatic patients rarely change management, and aren't indicated in many instances to begin with. And guess what? These patients usually have already received preop clearance from a cardiologist in the first place! So no, I don't care that you are fully dependent on a meaningless automated read.

What is interesting, however, is how quickly things change post op in the PACU when the patient now develops some symptoms. Still feel comfortable depending on the "non-specific ST/T wave abnormalities" read from the machine? I have several experiences from residency where anesthesia would consult much too late for an evolving STEMI that happened hrs ago (based on the initial EKG) but no one acted upon because the automated read said "normal sinus rhythm. normal ecg". As I said before, sensitivity is in the 60s for automated read detection. Its not what it finds, its what it doesn't.

I think most posters here are rational and we have a consensus on the importance of the human element in health care.

As physicians, we need to present a unified front against incurion of other interests.

 

[urge]

I figured you must be very early on in your training and cannot come up with a good discussion. Good luck in your future career through medicine!

 

[DrfluffyMD]

Fluff, I tend to agree that ai rads is probably further away than we think, but why do you think strong ai or agi is needed for diagnostics? Given a large enough normal variant / pathology database, clinical history, and sufficiently advanced image processing, shouldn't a machine be able to do say 99.5% of body ct or neuro?

Here's the thing, computer vision is good at pattern recognition, possibly one day greater than human, but image recognition is only one aspect of the human element.

Take an abdominal radiograph for example, a post op radiograph can have some free air, but how much of it is abnormal? That requires a complex decision tree which incoporates clinical status of the patient, white count, type of surgery and reading the op note (including reading between the lines about op note).

All of the above requires integration, higher order thinking and sometimes even imagnative skills.

The current state of machine learning requires building of "classifier" which categories the problem (if you can't identify the abnormal, you can't learn what it is). A lot of our work involves finding the problem.

Take the post op abdominal radiograph for example. If it's not implemented perfectly, surgeons are either gonna be swamped with fake abnormal free air reports or undercalls. Either way, they will no longer rely on the imaging result.

This is why an AGI is needed.

I had a write up somewhere about problem with AGI as well, in that you cannot ascertain the motivation of an AGI.

Such is also a problem with the current machine learning system. You don't know if the feature associated with atelectasis is actually the presence of a central line or not, because all the training set including atelectasis had central lines.

 

[Arch Guillotti]

Keep it professional or the thread will be closed. Thanks.

 

[Gastrapathy]

[QUOTE="DrfluffyMD, post: 18870045, member:

As physicians, we need to present a unified front against incurion of other interests.[/QUOTE]

Even if they do a better job?

 

[DrfluffyMD]

[QUOTE="DrfluffyMD, post: 18870045, member:

As physicians, we need to present a unified front against incurion of other interests.

Even if they do a better job?[/QUOTE]

I would gladly quit in an instant if computer can do my job better. I think the issue is that computer CANNOT do a better job for patient care for the reason I gave above.

The issue is that computer may do a "good enough" job in 98% of the time, and the power that be may decide that's good enough for patient care. I can see some analogy with CRNA in that. Ultimately, I believe patients are better served by an all MD anesthesia service, but the power that be decided that's not the case.

 

[deleted547339]

Here's the thing, computer vision is good at pattern recognition, possibly one day greater than human, but image recognition is only one aspect of the human element.

Take an abdominal radiograph for example, a post op radiograph can have some free air, but how much of it is abnormal? That requires a complex decision tree which incoporates clinical status of the patient, white count, type of surgery and reading the op note (including reading between the lines about op note).

All of the above requires integration, higher order thinking and sometimes even imagnative skills.

The current state of machine learning requires building of "classifier" which categories the problem (if you can't identify the abnormal, you can't learn what it is). A lot of our work involves finding the problem.

Take the post op abdominal radiograph for example. If it's not implemented perfectly, surgeons are either gonna be swamped with fake abnormal free air reports or undercalls. Either way, they will no longer rely on the imaging result.

This is why an AGI is needed.

I had a write up somewhere about problem with AGI as well, in that you cannot ascertain the motivation of an AGI.

Such is also a problem with the current machine learning system. You don't know if the feature associated with atelectasis is actually the presence of a central line or not, because all the training set including atelectasis had central lines.

I was right up there with you until you brought up the white count....

 

[WholeLottaGame7]

I love that when people try to tell us that nurses or computers can do our jobs as well as us, we (rightfully) give them all kinds of **** for not understanding the complexity of our field, but it seems like we have no qualms telling another medical specialty that a computer can do theirs.

People complain all the time about how physicians don't get any respect any more. Maybe it's because we can't even respect each other.

 

[urge]

We still don't have flying cars,...

http://nypost.com/2017/04/21/flying-car-successfully-completes-its-first-flight/

 

[RadsFTW123]

This gets thrown out there a bunch but to the best of my knowledge it's not true. I'm fairly certain you have to have a medical license in the US in order to get reimbursed for reads. Many places use nighthawk services but they are based in other parts of the US. Or like one radiology practice my wife knows, rents an apt in Paris and sends people over for a few weeks at a time to read overnight.

But it's not like overnight reads are being outsourced to Indian radiologists or something.

Agreed - it is illegal to bill Medicare unless you're US board certified and hold a state license in the state in which you are reading. Outsourcing is not happening. Could it happen? Sure... but we all know about the quality of foreign med schools and this has been technologically possible for a long time and doesn't happen.

 

[urge]

Agreed - it is illegal to bill Medicare unless you're US board certified and hold a state license in the state in which you are reading. Outsourcing is not happening. Could it happen? Sure... but we all know about the quality of foreign med schools and this has been technologically possible for a long time and doesn't happen.

Didn't we have a thread recently about FMGs having better outcomes that AMGs?

Tread carefully.

 

[pgg]

I love that when people try to tell us that nurses or computers can do our jobs as well as us, we (rightfully) give them all kinds of **** for not understanding the complexity of our field, but it seems like we have no qualms telling another medical specialty that a computer can do theirs.

People complain all the time about how physicians don't get any respect any more. Maybe it's because we can't even respect each other.

Not that long ago in my lifetime, there was reasonable doubt that computers would ever beat grandmasters at chess. Now chess is easily beaten, and computers are winning games of go against the best humans. If you know anything about the complexity of go, chess, and AI, you'd know that the search space for go is many orders of magnitude larger than chess. When Deep Blue beat Kasparov 20 years ago, the prevailing wisdom at the time was that go might never be solved. What's remarkable about it is that raw computing power didn't make it happen. It wasn't Moore's Law (alone), but rather algorithm advances.

The world is changing. AI will bring bigger changes to medicine than penicillin, or DNA sequencing, or anesthesia. There is exactly one field that is eminently vulnerable to learning machines, and that is diagnostic radiology, because 100% of their input data is 1s and 0s, and so is their output.

It's not disrespectful to anesthesiologists to admit that our field is threatened by CRNAs providing care that is "good enough" using the tools our predecessors and other scientists invented to make anesthesia safer and safer. It's not disrespectful to radiologists to admit that theirs is threatened by tools invented by other scientists.

I have great respect for radiologists. But even if I had an interest in doing it, I wouldn't pick the field unless my end goal was an interventional fellowship. (And not just because of the computer thing; the job market for radiologists who don't subspecialize sucks.)

I don't know when computers will be reading pap smears or chest x-rays and billing Medicare, but it'll happen in my lifetime.

 

[Stank811]

Dr Fluffy welcome to the anesthesia forum. Doom and gloom central of SDN. You get use to it overtime.

 

[WholeLottaGame7]

Not that long ago in my lifetime, there was reasonable doubt that computers would ever beat grandmasters at chess. Now chess is easily beaten, and computers are winning games of go against the best humans. If you know anything about the complexity of go, chess, and AI, you'd know that the search space for go is many orders of magnitude larger than chess. When Deep Blue beat Kasparov 20 years ago, the prevailing wisdom at the time was that go might never be solved. What's remarkable about it is that raw computing power didn't make it happen. It wasn't Moore's Law (alone), but rather algorithm advances.

The world is changing. AI will bring bigger changes to medicine than penicillin, or DNA sequencing, or anesthesia. There is exactly one field that is eminently vulnerable to learning machines, and that is diagnostic radiology, because 100% of their input data is 1s and 0s, and so is their output.

It's not disrespectful to anesthesiologists to admit that our field is threatened by CRNAs providing care that is "good enough" using the tools our predecessors and other scientists invented to make anesthesia safer and safer. It's not disrespectful to radiologists to admit that theirs is threatened by tools invented by other scientists.

I have great respect for radiologists. But even if I had an interest in doing it, I wouldn't pick the field unless my end goal was an interventional fellowship. (And not just because of the computer thing; the job market for radiologists who don't subspecialize sucks.)

I don't know when computers will be reading pap smears or chest x-rays and billing Medicare, but it'll happen in my lifetime.

I don't disagree with you in general about advances in AI, I'm just disagreeing with you about how soon, and who is most susceptible. I already stated that I can fully see a computer reading screening CXRs and mammos similar to screening EKGs, in our lifetime. And radiologists will be about as sad to see those go as cardiologists are to not read every single EKG. It's low-yield and tedious, which by definition makes it an easy target for our machine overlords.

But that doesn't mean diagnostic radiologists are going out of business any time soon. Radiologists already have a system of 1s and 0s, except they're called Hounsfield units. And the problem is, there are a lot of things that all look like a "1," and it requires context and critical thinking to figure out. So yeah, a computer might eventually do it, but not any time soon, and nobody on this board cares if a machine will be able to do their job in 2100. Just like nobody cares if a machine will be able to intubate and titrate vasoactive meds in 2100.

And I'm surprised by the push towards IR. Yeah they can do some weird stuff, but their bread and butter (lines, drains, etc) are ripe for midlevel encroachment. We talk on this forum all the time how procedural skills are monkey skills (which I tend to disagree with, but that's a different conversation), yet we're saying that's the safer of the 2 subspecialties?

The job market is definitely different. There appears to be a lot less turnover in radiology, and the average retirement age seems to be older (would be interesting to see this data by specialty), presumably because it's a less physically demanding field. Also, something like 90+% of residents complete a fellowship, so you are going to struggle to make yourself standout without one.

 

[urge]

It just so happens that radiology is the low hanging fruit of medicine.

 

[WholeLottaGame7]

It just so happens that radiology is the low hanging fruit of medicine.

I'm pretty sure that this forum has taught me that we're the low hanging fruit of medicine.

 

[nimbus]

I don't disagree with you in general about advances in AI, I'm just disagreeing with you about how soon, and who is most susceptible. I already stated that I can fully see a computer reading screening CXRs and mammos similar to screening EKGs, in our lifetime. And radiologists will be about as sad to see those go as cardiologists are to not read every single EKG. It's low-yield and tedious, which by definition makes it an easy target for our machine overlords.

But that doesn't mean diagnostic radiologists are going out of business any time soon. Radiologists already have a system of 1s and 0s, except they're called Hounsfield units. And the problem is, there are a lot of things that all look like a "1," and it requires context and critical thinking to figure out. So yeah, a computer might eventually do it, but not any time soon, and nobody on this board cares if a machine will be able to do their job in 2100. Just like nobody cares if a machine will be able to intubate and titrate vasoactive meds in 2100.

And I'm surprised by the push towards IR. Yeah they can do some weird stuff, but their bread and butter (lines, drains, etc) are ripe for midlevel encroachment. We talk on this forum all the time how procedural skills are monkey skills (which I tend to disagree with, but that's a different conversation), yet we're saying that's the safer of the 2 subspecialties?

The job market is definitely different. There appears to be a lot less turnover in radiology, and the average retirement age seems to be older (would be interesting to see this data by specialty), presumably because it's a less physically demanding field. Also, something like 90+% of residents complete a fellowship, so you are going to struggle to make yourself standout without one.

There's not a single radiologist (or orthopedist for that matter) on staff at my hospital who's not subspecialty fellowship trained. For now the radiologists are busy as f***. And the IR guys even busier.

 

[urge]

Not trying to be inflammatory, but everytime someone tell me that AI will take over radiology I tell them to look at sedasys or the effort to automate anesthesia. I think the effort was closer to reality versus AI in radiology. Yet, clinicians continue to prefer human being and the automation effort in anesthesia did not pan out.

The current state of art allow AI to learn how to catch a particular finding via a "classifer" that an AI scientist build. It's getting fairly good (not great, not diagnostically useful yet) to find things like a nodule on an X ray. However, AI is unable to make an independent diagnosis say, if you are trying to figure out if someone has a leak post whipple.

I have to say that as a second year, I hated radiology too. You just know enough to be toss to read a ton and don't know enough to feel competent. If I were you, I would stick it out for one more year and do IR. Most of IR now are NOT six hour cases. If you must switch though, you must switch.

Radiologists will go the way of the dinosaurs in our lifetime.

For a dollar, an AI will examine your medical scan

 

[anbuitachi]

Radiologists will go the way of the dinosaurs in our lifetime.

For a dollar, an AI will examine your medical scan

Seems like their way of improving their AI even further, by getting images from all over and having the machines learn from it.

 

[DrfluffyMD]

I say again, anesthesia delivery was ALREADY automated before radiology AI was made and guess what, people don’t like having a computer doctor.

It’s game over for the robot intended to replace anesthesiologists

Both anesthesia and radiology will be fine in my life time.

 

[Man o War]

AI only-read radiology is like traveling to a craphole country for cheap plastic surgery.....what could possibly go wrong.

 

[LLCoolK]

AI only-read radiology is like traveling to a craphole country for cheap plastic surgery.....what could possibly go wrong.

please explain how a computer reading images is like going to a craphole country for plastic surgery

 

[Man o War]

please explain how a computer reading images is like going to a craphole country for plastic surgery

Radiology on the cheap by cutting out the radiologist. You get what you pay for in both cases.

 

[deleted171991]

One doesn't need AI for radiology. One only needs cheap foreign docs (e.g. nighthawk), and a poor American sod to pin the malpractice responsibility on. Even worse than the firefighter concept from the Anesthesiology Care Team.

That's why diagnostic radiology has become such a tough field, that's why they end up doing at least one fellowship, that's why IR does much better.

People tend to focus on the wrong thing: total compensation. It's the unit compensation that matters, i.e. how much one makes per stress level and hour of work. Let's not forget also time off. That's why part-timers and mommy-trackers etc. tend to be happier people (and maybe even better paid, when adjusted for everything.)

 

[Man o War]

One doesn't need AI for radiology. One only needs cheap foreign docs (e.g. nighthawk), and a poor American sod to pin the malpractice responsibility on. Even worse than the firefighter concept from the Anesthesiology Care Team.

That's why diagnostic radiology has become such a tough field, that's why they end up doing at least one fellowship, that's why IR does much better.

People tend to focus on the wrong thing: total compensation. It's the unit compensation that matters, i.e. how much one makes per stress level and hour of work. Let's not forget also time off. That's why part-timers and mommy-trackers etc. tend to be happier people (and maybe even better paid, when adjusted for everything.)

Radiology is a lot like anesthesia IMO. People don't *really* understand what they do, they make it look easy, and therefore think they can replace those docs easily and cheaply. I have a ton of respect for those guys/gals, even though I'm obviously not a radiologist, I don't for a second think overseas people or AI can replicate their skills.

 

[deleted171991]

Radiology is a lot like anesthesia IMO. People don't *really* understand what they do, they make it look easy, and therefore think they can replace those docs easily and cheaply. I have a ton of respect for those guys/gals, even though I'm obviously not a radiologist, I don't for a second think overseas people or AI can replicate their skills.

Respectfully, that's ridiculous. Of course there are foreign docs who can diagnose (remotely) as competently as a local radiologist.

Any non-procedural specialty is at risk of being taken over by cheaper foreign docs (and local midlevels). The procedural ones by monkey see monkey do midlevels. There's the future for you, dear medical suckers students: rock, hard place, rock, hard place, rock...???

 

[AdmiralChz]

Respectfully, that's ridiculous. Of course there are foreign docs who can diagnose (remotely) as competently as a local radiologist.

Any non-procedural specialty is at risk of being taken over by cheaper foreign docs (and local midlevels). The procedural ones by monkey see monkey do midlevels. There's the future for you, dear medical suckers students: rock, hard place, rock, hard place, rock...???

This would be a great opportunity for the AMA to speak up and protect the profession of being a physician. Or address strongly the opiate crisis. Instead they are a weak organization focused on political healthcare changes.

 

[LLCoolK]

Radiology on the cheap by cutting out the radiologist. You get what you pay for in both cases.

but if there's data that shows that AI does a better job than radiologists for cheaper, you get even more than what you would have paid for hiring a radiologist

 

[anbuitachi]

I don't think AI replacing radiologists is even in the talks right now. Everything I've read has been about using ai as a tool for radiologists. That has to come first before radiologists can get replaced. Kind of like auto driving. First it's semi autonomous, then full autonomy. So we are no where close to AI replacing docs

 

[Man o War]

Respectfully, that's ridiculous. Of course there are foreign docs who can diagnose (remotely) as competently as a local radiologist.

Any non-procedural specialty is at risk of being taken over by cheaper foreign docs (and local midlevels). The procedural ones by monkey see monkey do midlevels. There's the future for you, dear medical suckers students: rock, hard place, rock, hard place, rock...???

Ok, then what is the standard? "Foreign doc" isn't good enough. There's a reason we make foreign docs do residencies in order to practice here, regardless of their previous experience.
They're like CRNAs, there is no standardization, so who knows what you're going to get.

 

[deleted171991]

Ok, then what is the standard? "Foreign doc" isn't good enough. There's a reason we make foreign docs do residencies here, regardless of their previous experience.

That's mostly protectionism. Otherwise, we would just let them sit for our board exams, especially for non-patient-facing specialties like radiology.

Even then, there are more and more US board-certified docs who return to their native countries, especially given our current friendly atmosphere. Many would rather work for 30-50% of their American salary "at home" and live even better than in the US. Mark my words; this is coming, more than AI. Anything that can be offshored/outsourced for much cheaper will be. It's already happening for overnight diagnostic radiology.

 

[DrfluffyMD]

That's mostly protectionism. Otherwise, we would just let them sit for our board exams, especially for non-patient-facing specialties like radiology.

Even then, there are more and more US board-certified docs who return to their native countries, especially given our current friendly atmosphere. Many would rather work for 30-50% of their American salary "at home" and live even better than in the US. Mark my words; this is coming, more than AI. Anything that can be offshored/outsourced for much cheaper will be. It's already happening for overnight diagnostic radiology.

With all due respect, are you a radiologist? You don’t seem to be very familiar with how radiology practice works before declaring outsourcing is taking over overnight rads

Because the whole offshore thing actually happened many years ago. A radiologists who is not in the US (even if trained in the US and boarded in the US) cannot bill for reads.

So there used to be outsourcing of overnight studies, where preliminary impressions are made (not billed) and the actual studies are read in the morning. The radiologists working during the day pay for the priviledge of not having to work at night.

As the job market worsened a few years ago, practices can suddenly afford actual night time local coverage and can no longer afford paying people to prelim studies which is a net cost (since you make less now with those studies). The whole overnight teleradiology industry then suffered.

Now, as radiology job markets get better, some rads can afford to make lifestyle related demands like night time teleradiology coverage and use that as recruitment tool. The night time preliminary read business exist to make rads’s life better, not replace them.