IBM Watson replacing radiology eventually?

This forum made possible through the generous support of SDN members, donors, and sponsors. Thank you.
Ischemia1032

Ischemia1032

Full Member
7+ Year Member
Joined
Jul 29, 2015
Messages
87
Reaction score
16
Last edited:
If the skills of Dr. EKG-Machine fortell anything here...

This is never going to happen. The amount of times ecg machines have misinterpreted a 12 lead are absolutely insane (calles a 4-1 flutter a fib, or calls q waves all over the place). This is a whole lot simpler (voltage criteria ect..) than complicated imaging modalities...

The nuances are infinite imo.
 
Last edited:
lol sorry: I fixed it, the correct link is up... and the reality is, at least I think based on my own opinion and speaking with other attendings, that if a company(like IBM) could prove that their supercomputer could read with the same efficacy as a radiologist, then they could at least start by being put in places like underserved areas etc., people may not trust it, the question is will congress trust it enough to pass legislation that allows them to use it instead of radiologists.
 
Psai said:
No one would trust a machine's reading. And who would take the responsibility for a bad outcome?
Click to expand...

People trust electronic blood pressure cuffs to diagnose them with hypertension, no one requests that a doctor take their blood pressure reading manually because they don't trust the machine. I know radiology is much more complex than blood pressure reading, but I don't think most people consider complexity.

As for liability, the company that makes the machine, also has to deal with malpractice. IBM and other huge corporations are better equipped to fight/absorb/deflect liability than a hospital or physician.

All that said, I imagine we've still got a while. I have to assume there are simpler jobs than radiology that will be automated first.
 
Ischemia1032 said:
lol sorry: I fixed it, the correct link is up... and the reality is, at least I think based on my own opinion and speaking with other attendings, that if a company(like IBM) could prove that their supercomputer could read with the same efficacy as a radiologist, then they could at least start by being put in places like underserved areas etc., people may not trust it, the question is will congress trust it enough to pass legislation that allows them to use it instead of radiologists.
Click to expand...

So you want to start experimenting on the poor and those who can't help themselves? Can't think of any time that has gone poorly in the past.

justabanana said:
People trust electronic blood pressure cuffs to diagnose them with hypertension, no one requests that a doctor take their blood pressure reading manually because they don't trust the machine. I know radiology is much more complex than blood pressure reading, but I don't think most people consider complexity.

As for liability, the company that makes the machine, also has to deal with malpractice. IBM and other huge corporations are better equipped to fight/absorb/deflect liability than a hospital or physician.

All that said, I imagine we've still got a while. I have to assume there are simpler jobs than radiology that will be automated first.
Click to expand...

The bolded literally happens all. the. time.
 
If you are worried about this, you should not go into medicine.

Sent from my XT1092 using Tapatalk
 
Psai said:
No one would trust a machine's reading. And who would take the responsibility for a bad outcome?
Click to expand...
Right now this is the case, but in a generation or two where the general population is used to everything being automated? I wouldn't be so sure.
 
justabanana said:
People trust electronic blood pressure cuffs to diagnose them with hypertension, no one requests that a doctor take their blood pressure reading manually because they don't trust the machine. I know radiology is much more complex than blood pressure reading, but I don't think most people consider complexity.

As for liability, the company that makes the machine, also has to deal with malpractice. IBM and other huge corporations are better equipped to fight/absorb/deflect liability than a hospital or physician.

All that said, I imagine we've still got a while. I have to assume there are simpler jobs than radiology that will be automated first.
Click to expand...
The machine would have to be as accurate as physicians to make this worthwhile, as IBM certainly wouldn't want the risk that comes with an unreliable machine that functions as a lawsuit generator otherwise.

I mean, if we can't get EKGs right, Watson is a long ways off...
 
justabanana said:
People trust electronic blood pressure cuffs to diagnose them with hypertension, no one requests that a doctor take their blood pressure reading manually because they don't trust the machine. I know radiology is much more complex than blood pressure reading, but I don't think most people consider complexity.

As for liability, the company that makes the machine, also has to deal with malpractice. IBM and other huge corporations are better equipped to fight/absorb/deflect liability than a hospital or physician.

All that said, I imagine we've still got a while. I have to assume there are simpler jobs than radiology that will be automated first.
Click to expand...

Since when? I've never seen a doctor diagnose someone with hypertension without checking it manually first. Even in the hospital I am asked frequently to check blood pressure manually to confirm
 
Let's look where computer assisted detection is already used in Radiology...Mammography. It is way too sensitive and you ignore 95% of what is says. If you biopsied everything it marked, the breast would be Swiss cheese. And this is simple 4 view radiographs.
 
DrBowtie said:
Let's look where computer assisted detection is already used in Radiology...Mammography. It is way too sensitive and you ignore 95% of what is says. If you biopsied everything it marked, the breast would be Swiss cheese. And this is simple 4 view radiographs.
Click to expand...

But if it was sensitive you would have high true positives
 
http://xkcd.com/1425/

People don't understand the nuance it takes to properly read imaging. By the time computers take over a radiologist's job they will have taken over most other jobs as well.
 
I agree how nuanced it is, people always look at radiology, and many of the residents I have met think it is an easy residency. Radiology requires much more studying then almost any other medical specialty(with the exception of pathology which is probably equal if not more). The reason radiologists will say legion may or may not be present, clinically correlate. Is because radiologists are afraid of making a definitive diagnosis because it makes them very liable, with the amount of scans they read, they fear getting sued. Radiologists here are very well trained and know what they are doing, but many are afraid. The best radiology attendings are the most cavalier in my experience.
 
Ischemia1032 said:
I agree how nuanced it is, people always look at radiology, and many of the residents I have met think it is an easy residency. Radiology requires much more studying then almost any other medical specialty(with the exception of pathology which is probably equal if not more). The reason radiologists will say legion may or may not be present, clinically correlate. Is because radiologists are afraid of making a definitive diagnosis because it makes them very liable, with the amount of scans they read, they fear getting sued. Radiologists here are very well trained and know what they are doing, but many are afraid. The best radiology attendings are the most cavalier in my experience.
Click to expand...

Oh god. Everyone knows why it says clinically correlate, now stop with the asinine comments that radiologyrequires much more studying than any other specialty
 
Psai said:
Since when? I've never seen a doctor diagnose someone with hypertension without checking it manually first. Even in the hospital I am asked frequently to check blood pressure manually to confirm
Click to expand...
I believe it. My lack of clinical experience is showing. I never saw anyone ask for a re-do in the ER, but I suppose I've also never seen a diagnosis of hypertension there either where the patient wasn't aware of it. I was just trying to go for an example where people trust machines more than they reasonably should. There are a lot of example outside of medicine (GPS..), but I can't think of any in medicine if even electronic blood pressure readings are then done manually before making diagnoses.

Regardless of whether people would trust a computer to make a medical diagnosis, I definitely think a lot of the simpler jobs will be automated long before radiology.
 
JP2740 said:
Oh god. Everyone knows why it says clinically correlate, now stop with the asinine comments that radiologyrequires much more studying than any other specialty
Click to expand...
I totally disagree on your point that radiology requires equal studying to other medical specialties. Radiology is required to know a lot about everything, you need to be perfectly competent in every part of the body, with more diseases then any one doctor(once again with the exception of pathology, I consider them a field similar to us). I am not saying it is more difficult, I am just saying it requires much more studying as is in the nature. You are staring and diagnosing conditions all day based on pictures, each case presents in many different ways, and I know what you are going to say. IM for example, a case can present in 10 different ways. In radiology one diagnosis can present in 50 different ways. The skill it takes givent he the time constraints of imaging is high. It simply requires a much larger basic science knowledge base then most clinical specialties. Clinical specialties(while radiology is still technically clinical, but I don't consider it on the same level as IM for example) such as IM and it's subspecialties require much more practicing different clinical cases and reading about cases, then actual studying of new information consistently, radiologists not only have to learn a totally new way to process medical diagnostics, but they have to keep up reviewing new anatomy. For most clinical specialties, it is not going to matter if you forgot some neuroanatomy, or cardiothoracic anatomy here and there since they are trained in a different way, but to a radiologist, anatomy is everything. Constant studying the entire body, it is not easy by any means, I do not think it is necessarily harder overall, I do think radiology is more cognitive then internal medicine based on my experiences, not to say internal medicine isn't cognitive, because it is, I think it requires a lot more social thinking then radiology which means it's duties are more split, it is just difficult in different ways, just like how surgery is difficult in different ways then medicine. Why do we always have to make it a competition, all specialties require different goals, some require more studying then others, others require more case review, others require more practice of techniques, who cares honestly, why put eachother down?
 
Ischemia1032 said:
I totally disagree on your point that radiology requires equal studying to other medical specialties. Radiology is required to know a lot about everything, you need to be perfectly competent in every part of the body, with more diseases then any one doctor(once again with the exception of pathology, I consider them a field similar to us). I am not saying it is more difficult, I am just saying it requires much more studying as is in the nature. You are staring and diagnosing conditions all day based on pictures, each case presents in many different ways, and I know what you are going to say. IM for example, a case can present in 10 different ways. In radiology one diagnosis can present in 50 different ways. The skill it takes givent he the time constraints of imaging is high. It simply requires a much larger basic science knowledge base then most clinical specialties. Clinical specialties(while radiology is still technically clinical, but I don't consider it on the same level as IM for example) such as IM and it's subspecialties require much more practicing different clinical cases and reading about cases, then actual studying of new information consistently, radiologists not only have to learn a totally new way to process medical diagnostics, but they have to keep up reviewing new anatomy. For most clinical specialties, it is not going to matter if you forgot some neuroanatomy, or cardiothoracic anatomy here and there since they are trained in a different way, but to a radiologist, anatomy is everything. Constant studying the entire body, it is not easy by any means, I do not think it is necessarily harder overall, I do think radiology is more cognitive then internal medicine based on my experiences, not to say internal medicine isn't cognitive, because it is, I think it requires a lot more social thinking then radiology which means it's duties are more split, it is just difficult in different ways, just like how surgery is difficult in different ways then medicine. Why do we always have to make it a competition, all specialties require different goals, some require more studying then others, others require more case review, others require more practice of techniques, who cares honestly, why put eachother down?
Click to expand...

TLDR?

You're jl linning this **** up.
 
justabanana said:
I believe it. My lack of clinical experience is showing. I never saw anyone ask for a re-do in the ER, but I suppose I've also never seen a diagnosis of hypertension there either where the patient wasn't aware of it. I was just trying to go for an example where people trust machines more than they reasonably should. There are a lot of example outside of medicine (GPS..), but I can't think of any in medicine if even electronic blood pressure readings are then done manually before making diagnoses.

Regardless of whether people would trust a computer to make a medical diagnosis, I definitely think a lot of the simpler jobs will be automated long before radiology.
Click to expand...

? Why would the er be diagnosing people with hypertension? They don't even treat high blood pressures usually unless it is something like tpa or hypertensive emergency
 
I think that a big part of the problem with making computers that do this sort of work is structural. A lot of people think that it's just a matter of time until artificial intelligence reaches or surpasses the ability of the human mind to do every sort of operation, but I don't think this is the case—at least not until we fundamentally change the way we approach computers.

Computers are very good at formal logical operations and deductive reasoning. The fundamental problem here is one that has long been recognized by logicians: formal logic actually does a very poor job of describing the world except in select concrete arenas. Formal logic is very good for elucidating explicit, technical meaning within well-defined, pre-established parameters. It does not help at all with the clarification of the myriad problems where limits are not preset and well-defined (i.e. Fuzzy Logic or abductive logic). For instance: Computers can be excellent if the task is basically to be a dictionary—say, "What is the color blue." They do much poorer jobs as the tasks become more abstract and require more abductive reasoning (reasoning from observations back to the most parsimonious unifying theoretical construct) and less deductive reasoning—for example, "What is meant by blue in the context of this poem?" or even more extreme "what is the meaning of blueness?"

Most of medicine is extremely abductive by nature. One must contextualize all sorts of findings and information in order to form theories that explain them and then compare the available theories in terms of probability/parsimony. Many of the parameters that enter into the ability of a human to perform this operation are gained through experience as a living being interacting with the world and are difficult to enter and weight as strict variables within a computer program. The weight of any variable can change drastically depending on context. Most attempts to make computers that can do this type of operation well have approached it as trying to basically "trick" a formal, deductive machine to approximate abductive reasoning by inputing a complicated framework of variables and rules that the machine can follow. I think that in real application, these machines will quickly hit ceilings based on their inability to independently approximate context and intuitively modify the algorithm to achieve the best possibly answer.

Of course some computers have attempted to use "learning" approaches to artificial intelligence and I think these will be more likely to effectively do the sort of operation we're talking about here. I'm unsure whether they will soon or ever do so in reality, but it is a more promising approach.

Then again, I'm not a programmer or an expert on artificial intelligence so I might have missed some things. This is just my philosophical evaluation based on what I have heard about computer science and AI.
 
Ischemia1032 said:
What do you all think, I found a very interesting article, it seems to be going pretty far. So far that the radiologists I know are recommending to not go into radiology unless you are on a DR/IR combined residency because they are feeling if radiology can be taken over, eventually so will the rest of medicine and most of medicine will consist of procedure based specialties.

http://www.forbes.com/sites/realspin/2016/02/11/automation-telemedicine/#25056fbc5b97
Click to expand...

The author is a radiologist, of course he's going to jump straight from IBM bought Merge Healthcare (http://www-03.ibm.com/press/us/en/pressrelease/47435.wss) to radiologists are out of a job. If you read what IBM plans do with these scans, it's more likely than not going to save a ton of lives. Why speak out against what makes people better? We have clinical research for this type of stuff.
 
Pssht. someone needs to bring shark200o in here to school everyone
im not tagging, cause i don't wanna bug him. He's responded a billion times to claims like this.
 
You must log in or register to reply here.

Similar threads

@
Torn: Orthopedics vs Radiology
Replies
20
Views
3K
@#Frogthejam#@
@
M
ACR Radiology Summer Research Internship
Replies
4
Views
1K
medstudent_29
M
M
Interventional Radiology Electives Advice
Replies
0
Views
952
mshalom13
M
N
Ortho vs Radiology
Replies
36
Views
8K
Cognovi
Cognovi
A
How should I network at interventional radiology conference (SIR 2025)
Replies
0
Views
816
AHopefulApplicant
A
Top