-
How will AI affect radiology over the next 20 years?
[link=http://www.auntminnie.com/index.aspx?sec=sup&sub=aic&pag=dis&ItemID=117460]http://www.auntminnie.com…&pag=dis&ItemID=117460[/link]
“Within 15 to 20 years, deep-learning algorithms will be able to produce reports for most diagnostic imaging studies, Erickson said.”
I think Erickson is too conservative. I believe AI will be able to produce reports within 10, maybe even 5 years.
I think Siegel is beginning to understand.
-
373 Replies
-
Sounds like the exact same talk at RSNA last year.
-
[link=http://doctordalai.blogspot.com/2016/12/artificial-intelligence-at-rsna-im.html]http://doctordalai.blogsp…igence-at-rsna-im.html[/link]
I stand by every word.-
I remember reading a book about medical specialty selection in 1994 that predicted all radiology reports would be generated by AI within 5 years of publication, or maximum 10 years, so only idiots would go into radiology.
Ah those were the days.
-
They can’t even get computers to read mammograms successfully.
Keep in mind that a CT study is essentially a 4 dimensional study (x, y, z coordinates plus attenuation values) it’s not going to be as simple as running a million 3D images to train a neural network. Even mammograms and radiographs are 3 dimensional studies. Luckily a lot of radio-logic exams have artifacts.
They should concentrate on something easier in the short term. How about a robot that can accurately protocol all studies based on appropriateness? I think the robot would malfunction when it sees the indication of PE vs chole vs appy vs renal stone-
Don’t change your prediction now Jimboboy. You said 10 years when I was a PGY 3. That was 2 years ago, 8 years more to radiologist-replacing robots. I’m marking your words as I’m taking online classes for an MBA due to fears of job replacement.
-
I might mischievously point out that this emerging technology is ripe for… ahem… career-prolonging “interventions” (wink, wink) by “contributing” radiologists.
-
Hear Ye
Hear Ye
Somebody has to teach the machines…
Might as well be us.
[;)]-
You and I gave it teaching material already. Your reports and accompanying imaging studies are all it needs.
-
Fast forwarding 10 years from now….
Impression:
1. Right pneumothorax vs bleb vs artifact, correlate clinically.
2. Aortic dissection vs plaque vs artifact, correlate clinically.
3. Free air vs bowel gas vs artifact, correlate clinically.
I guess the clinician is left to figure out all this by himself since all radiologists have been replaced by AI. Good luck!
-
-
-
Eight years huh…..I bet if FDA clearance wasn’t so slow they could do it in 2 years with various algorithms that monitor the state of every every signal on a flat panel detector and reference that study with 1,000’s of “like” stored results that provided a wide data base for various anomalies like weight of the patient, previous recent scans or x-ray results of the patient and a database of 100,000 of thousands that are her very close match in all facets including lifestyle habits…..with the monitoring of the digital matrix lines of a flat panel detector and the ciphering of possibly millions of other patients with know results from a previous database I spoke of I believe a computer diagnosis could be the real deal quite quickly. ( Just 1’s and 0’s and some “in between’s”) Except it takes us humans forever to get the computer what it needs to do it’s job! Haha……yea, so maybe a bit off yet!
[i][b]Rapid X-Ray – Digital X-Ray Sales & Service[/b][/i]
[i][b][size=”2″]”The bitterness of poor quality remains long after the sweetness of low price is forgotten”[/size]
[/b][/i]
[color=”#0000ff”][u][link=https://www.rapidxray.biz]https://www.rapidxray.biz[/link][/u][/color]
[color=”#0000ff”][u]
[/u][/color]
-
-
-
-
-
-
It’ll be an explosion in false positives.
-
A “radiology is doomed by AI” thread from JimboBoy? Say it isn’t so!
-
and here we go again.
some talking head makes a doomsday prediction that they can in no way defend – just expression of opinion often by someone who knows nothing about radiology or medicine (Like Obama). Then we all jump in to say how stupid and uninformed he is. Repeat every month.
This seems like a good application for structured reporting – simply make a structured report to respond to this kind of post, adding in any details (like a name) that are slightly different, push a button and post. That structured report software we were all supposed to be using now isn’t doing anything else these days. May as well be used to provide a standard reply to these posts.
On to better things……..-
BTW jimbobob:
Your posts are so relentless on this topic, it would be good information to be able to understand where you are coming from if you could let us know what your position in the radiologic social structure is: Tech? Radiologist? admin? researcher in AI? -
@jimboboy
Will AI companies have legal rights to use reports with accompanying images? Are those public domain?
Another question: if it gets to the point where companies buy image/report packages from groups, would radiologists have the strength to tell them to go to hell?
-
The answer is for us to coordinate efforts and GET CONTROL of this before some other entity does. I’m still not worried about getting replaced by Watson, but this technology could help US if we could get into the driver’s seat and keep others out. The ACR is slowly moving in that direction. Finally they’ve done something of which I wholeheartedly approve.
-
Jan, when I said “you and I”, I meant it in a generic sense.
To be more specific, AI can be fed studies and accompanying reports in a participating network. For example, imaging studies done in at practices that use MERGE can be sent to train Watson. In fact this is probably happening as we speak. I should mention that these studies are likely being sent without patient identifiers, so there is no need for consent. This last part answers your second question.
-
Not to quibble, but while the first part of your post addresses my last question, the last part does not. Patient privacy isn’t the issue; my question was whether AI companies have rights to reports generated by radiologists who never consented to the dissemination of their work.
-
This is an example of a situation in which radiologists are at the mercy fo their corporate masters, who are more interested in 10-15 years of payoffs from AI companies than they are in the long-term viability of the specialty. If Rads dies, the leeches will join their comrades in primary care specialties, surgical fields, etc.
-
Jan,
your report belongs to your boss (aka partnership, imaging center, hospital, university, corporation, etc). The “boss” in turn has contractual relationships with various business arrangements, including the likes of Merge. Ownership of your report left you when you uttered the dictation. You have no rights to it.-
Jimboboy:
Who are you? Please let us know
-
-
-
-
-
-
-
-
-
The details of my life are quite inconsequential
very well, where do I begin? My father was a relentlessly self-improving boulangerie owner from Belgium with low grade narcolepsy and a penchant for buggery. My mother was a fifteen year old French prostitute named Chloe with webbed feet. My father would womanize, he would drink. He would make outrageous claims like he invented the question mark. Sometimes he would accuse chestnuts of being lazy. The sort of general malaise that only the genius possess and the insane lament. My childhood was typical. Summers in Rangoon, luge lessons. In the spring wed make meat helmets. When I was insolent I was placed in a burlap bag and beaten with reeds- pretty standard really. At the age of twelve I received my first scribe. At the age of fourteen a Zoroastrian named Vilma ritualistically shaved my head. There really is nothing like a shorn head its breathtaking- I highly suggest you try it.
-
Sardonicus, I’ve stated on more than one occassion that I’m a practicing radiologist.
-
Jimboboy, you have posted rather relentlessly about AI taking over. You DO realize you are in the extreme minority on this, and that you are scaring a lot of the children, I mean medical students, who read what I can only call your fear-mongering and think that radiology is on its death-bed.
WHAT is your deal? Do you own IBM stock? Any family history of paranoia? [;)]-
Great Austin Powers reference. Did you memorize it? Or cut and paste?
-
If you’re a radiologist and want to hedge against AI taking away your livelihood you could always start investing in the [url=https://www.globalxfunds.com/funds/botz/] GlobalX
[h1]Robotics & Artificial Intelligence ETF[/url], [BOTZ].[/h1]
It’s close to a pure AI play. 🙂
-
To hedge you’d need to plunk down 1/5 to 1/3 of your salary for 5 years to build up enough principal in it. Chances are , earnings there will not replace your lost income, unless you have a mil plus
-
AI will ABSOLUTELY replace radiologists, assuming you have a realistic definition of “replace”. There is ZERO debate about that. Consider this: did PACS replace radiologists?
YES! It replaced tons of radiologists because it made us a lot more efficient. If you now need 3 rads when you used to need 4, well then 1 got “replaced” by PACS. AI will do the same thing — pre-generate reports, prioritize worklists, pre-annotate images, etc. You will read a lot of studies like people read DEXA (often just one click). It will make folks incredibly more efficient.
Do I think a rad will be eliminated from the diagnostic equation? I hope not. Presumably, just like the computer-prelim’ed EKGs are overread by a cardiologist, we will do the same. But it is certainly possible that some enterprising non-radiology physician will just say “the AI is good enough for me — I’ll take the risk and sign off on all those AI-prelim’ed reports and make that cash.” Or maybe a hospital system decides that a handful of Rad-PAs checking the AI-reports is fine, with one actual rad overseeing the RPAs.
You might have a very myopic definition of “replacement”– images are fed to the AI and the AI reports back to the referring physician, and no intervening human is required. However, if you have a more real-world definition of replacement–AI will lead to needing half as many radiologists as we used to have–then certainly it becomes more of a certainty that AI will replace A LOT of rads.-
A great argument above for reduction of training spots.
-
-
-
-
Quote from NYC
Great Austin Powers reference. Did you memorize it? Or cut and paste?
[link]https://www.youtube.com/watch?v=ZO3pUVbNSnA[/link]
-
Why does this topic keep resurrecting itself on this forum? By the time our software and hardware is capable of replacing radiologists it also will be able to replace most professions. This will change our society.
-
-
-
-
-
-
It keeps recurring because Dr. Jimboboy creates a new thread every time he finds an article that supports his pessimistic view. This has become about as futile as the Right vs. Left discussions on the Political board. However, unlike the rabid, impotent hate-mongering that goes on over there, this is a topic wherein we [i]could[/i] have an impact; we can do whatever we can to steer this new technology, rather than let someone else do it. But to sit in sackcloth and ashes, and whine about our imminent demise does no one any good, save perhaps for those who own stock in AI companies, and who think they might benefit by scaring med students away from radiology.
-
Put your time in, become financially independent and hope you aren’t a debt slave once you’re forced to start working for WalMart.
-
zz111 is exactly right.
AI will make radiologists more efficient to the point that fewer of us are needed (and this will happen in next 5-20).
Full replacement will take (a lot?) longer.
Current state of the art achieves near human performance on screening mammography (specificity 0.8 at sensitivity of 0.8, humans are around 0.9 at similar sensitivity) and will get better.
Anyone using examples (e.g.old fashioned CAD) from before 2012 (when convolutional neural networks first won ImageNet) is comparing apples to oranges.
-
Quote from anonirad2012
Current state of the art achieves near human performance on screening mammography (specificity 0.8 at sensitivity of 0.8, humans are around 0.9 at similar sensitivity) and will get better.
reference please. Need to review materials and methods.
-
Competition, ongoing, legit, representative large dataset.
Final results and publication pending.
-
[link]https://www.synapse.org/#!Synapse:syn4224222/wiki/[/link]
-
Interesting. The DREAM competitors seem to be primarily from places other than the US. Also, the competition was sponsored by the Laura and John Arnold Foundation…
[link]http://www.arnoldfoundation.org/about/[/link]
[blockquote] Our core objective is to improve the lives of individuals by strengthening our social, governmental, and economic systems.
[size=”0″]Strategy[/size]
Our strategy is to systematically examine areas of society in which underperformance, inefficiency, concentrated power, lack of information, lack of accountability, lack of transparency, lack of balance among interests, or other barriers to human progress and achievement exist. We then apply a rigorous and comprehensive entrepreneurial problem-solving approach to these areas, considering all possible strategies, tactics, and resource allocations to effect solutions. Our approach is not limited to what has been tried, or even what has been proposed, in the past. Instead, we seek to incentivize bold, creative thinking and effort, with the goal of igniting a renaissance of new ideas and approaches applied to persistent problems.
[/blockquote] Read into it what you will. Being paranoid, I sense folks who don’t like doctors.
-
-
-
Dalai, can you tell me again who is it that tends toward conspiracies? Can you get over your theories about me trying to scare med students and having financial motives?
I started posting about AI 5 years ago (google AI learning to see cats on its own) on Auntminnie because I understood what it meant for our profession. I saw how it was nothing like traditional mammo CAD and that the technology was but a few steps away from medical image interpretation. You may not realize it, but the time for discussions about how us radiologists altering that course has passed.
I really don’t understand how you don’t think this is (“was” is more appropriate) an important topic. THIS IS OUR LIVELIHOOD. Even Auntminnie has an article about this topic every other day. But you’d rather avoid the subject and pretend it will never happen, wouldn’t you?
-
Do I have to start altering your name again? I just love it when I get called on the carpet for my bad behavior, but…
So since you are the expert in the future of our field, please tell us po’ folk just what we are supposed to do… You have been rending your garments, and sitting in sackcloth and ashes (biblical references…the atheists won’t get it) and preaching about our upcoming doom and devastation. So?
JUST WHAT IS IT YOU WANT TO ACCOMPLISH? You (and only you) understand “what it meant for our profession”. And now it’s “too late to talk about it,” according to you.
No one said it isn’t (“WASN’T”) an important topic… But let me tell you something, sonny…Your constant bleating gets us nowhere. [i]I’ve[/i] spoken to Dr. Siegel and asked how to proceed. [i]I’ve[/i] talked to the muckety-mucks from IBM about Watson. But [i]your[/i] contribution is to find every single fear-mongering shred of information you can and slap it up here as quick as you can so you can continue to pi$$ and moan about our future. Again, you are accomplishing nothing more than scaring the children. Or are you like some sort of sit-com mother, getting secondary gain by making sure you are the one who tells everyone the sky is falling?-
So what was Siegel’s answer? That you shouldn’t worry because it won’t happen in our lifetime? How is that useful if he is wrong?
And what was IBM’s answer? That it is only here to help, not take over? Again how is that helpful if your job gets taken over.
My aim is to inform so you can prepare, whatever that means for you. It may mean seeking additional skills. It may mean building wealth. But whatever you do, don’t assume that your job is secure for next 2-3 decade.
I know that is not what you want to hear from me. You want me to say “everything will be fine. You have nothing to worry about, AI is hear to help us, not replace us”
-
So how are you preparing for this calamity? What other skills are you acquiring and how are you managing your wealth any better? Let’s hear it from you.
-
For what it’s worth, one of the interns who is rotating with us, and is going into radiology next year in another state said he’s contemplating switching into an open position in EM at his home program – as he’s tired of hearing about AI taking over jobs and it’s a risk he doesn’t want to take.
My younger brother is an M2 and he says that every other kid at his school mimics what’s posted on here about artificial intelligence replacing radiologists and it being screwed as a field – that only IR is worth going into as procedures are a safe buffer.
If you guys don’t think medical students and interns don’t read this ,you’re wrong. I don’t get what the point of all the fear mongering is. How realistic do you think it is that a machine is going to literally replace you and do your job and you will be out on the street? Please post logical arguments.
Will there be less jobs? Maybe / Probably. Will money go down ? Probably – as it will in every other specialty. Are we going to be replaced by a supercomputer and have no job? Impossible. Get real guys.
This happened in Anesthesiology with the insane numbers of CRNA’s that have been churned out over the past 20 years – Are jobs harder to get for anesthesiologists as a result? Sure. Do they spend a lot more time covering more cases because they are monitoring CRNA’s? Absolutely .Have they, or will they be replaced, as SO many old timers cried about back in the 90’s when this all stared? Absolutely not. Has income gone down? Yes, but in proportion to everyone else.
I don’t see this as being much different in radiology. BUT, as a resident who also sits here and reads everything you’ve got to say, I will say that it’s scared me the heck away from doing my passion – which is neuro rad. I’ve talked to a few of our seniors that applied this year and they’ve echoed their fears and it’s impacted what fellowships they are applying to. Some have shied away from Body and Neuro as a result – as people say these are likely the ‘easier’ fields for a machine algorithm to take over..Our seniors are exclusively applying to Breast and IR as these are seen as fields where you ‘need’ a person doing procedures. It seems nuts for me to make such a big decision simply out of fear of the ‘What if” .
That being said, i’m the biggest hypocrite of all – I’ll probably end up doing IR now just so i have a ‘skill set’ that isn’t automatable. I know that means a likely miserable lifestyle and turf wars forever, but hey, until a machine can de-clot a fistula, put in a line, drain some abscess and take some biopsies, ( whatever is left over that vascular, cards, nephrology, etc doesn’t steal in the next 5 years ) at least It’s one less thing for me to worry about.
Do i really believe that diagnostics will get replaced by machines? No. Just being a mid level resident and seeing the complexity in what an average radiologist does day to day makes me think that’s unlikely. It’s easy for a tech person who doesn’t know what the heck they’re talking about to look at us and say “Oh, your job is all about repetition and algorithmic thinking. I can teach a machine to do that.” – But it’s an entirely different thing when so many experienced Attending physicians on this board echo that statement with such bold ness.
After reading post after post after post, fear has crept in. I’m young. I saw this awful radiology job market hit 5 years ago and it’s slow recovery, I don’t think it’s an invincible field by any means. I have 30 years of practice ahead of me. My wife and i are BURIED in loans. We are both working our tails off. I’d like to buy a house one day. Or a car. Or take a freaking vacation that isn’t to a waterpark. I can’t afford for my skills to become worthless in the next 10 years, if that’s what so many here are predicting. This thread has been truly such a bummer for my weekend..
-
Are you happy now, Jimboboy? Your [strike]trolling[/strike] posts do have some very negative effects. Congratulations, sonny. You’ve scared the children, and why you insist on continuing to do so is beyond me. The AI discussion is one we need to have, and I actually appreciate your bringing some esoteric stuff to our attention. But when you do so, it is only in the context of how much more imminent the falling sky has become. By the way, the forum does have a blocking feature. Feel free to block me. I shall not be returning the favor, as obviously, your doom and gloom posts need to be countered.
I’ve seen two legitimate scenarios for AI. One is for it to serve as a super CAD…to [i]assist[/i] us in reading. That we all should welcome, assuming it actually improves our ability to deliver patient care. The second is as a screener of sorts, to be applied in underserved areas. Having done a mission trip to such an area in Africa (have YOU considered giving back in this manner, sonny?) and by and large, they do not have the infrastructure to support AI as it stands today. There is a huge struggle to even bring PACS to these areas, and bandwidth is so dear there is no chance of supporting routine remote reads, human or AI, for the foreseeable future. (If anyone cares about such things, I strongly urge you to donate to, or better yet volunteer with Rad-Aid.)
So, instead of preaching imminent doom, GO DO SOMETHING!
And Rads312, don’t let this discourage you. You will be doing whatever it is you choose to do for the majority of your waking hours for the next 30+ years. Do whatever you love, whatever you have a passion to do. I went with Nuclear Radiology because it fit me well, and I love practicing it. It severely limited the number of jobs, even back in the bad ol’ days when jobs were more plentiful. I’ve never regretted the choice. You don’t want to look back when you are 65 and realize you really hated what you did all those years. Don’t sell your soul for false promises and empty guarantees. Who knows? The surgeons might take all IR away, and I’m more wary of them (and administrators) than any pile of transistors! Now [i]that’s[/i] fear-mongering!!!! [:)] -
Quote from rads312
For what it’s worth, one of the interns who is rotating with us, and is going into radiology next year in another state said he’s contemplating switching into an open position in EM at his home program – as he’s tired of hearing about AI taking over jobs and it’s a risk he doesn’t want to take.
My younger brother is an M2 and he says that every other kid at his school mimics what’s posted on here about artificial intelligence replacing radiologists and it being screwed as a field – that only IR is worth going into as procedures are a safe buffer.
If you guys don’t think medical students and interns don’t read this ,you’re wrong. I don’t get what the point of all the fear mongering is. How realistic do you think it is that a machine is going to literally replace you and do your job and you will be out on the street? Please post logical arguments.
Will there be less jobs? Maybe / Probably. Will money go down ? Probably – as it will in every other specialty. Are we going to be replaced by a supercomputer and have no job? Impossible. Get real guys.
This happened in Anesthesiology with the insane numbers of CRNA’s that have been churned out over the past 20 years – Are jobs harder to get for anesthesiologists as a result? Sure. Do they spend a lot more time covering more cases because they are monitoring CRNA’s? Absolutely .Have they, or will they be replaced, as SO many old timers cried about back in the 90’s when this all stared? Absolutely not. Has income gone down? Yes, but in proportion to everyone else.
I don’t see this as being much different in radiology. BUT, as a resident who also sits here and reads everything you’ve got to say, I will say that it’s scared me the heck away from doing my passion – which is neuro rad. I’ve talked to a few of our seniors that applied this year and they’ve echoed their fears and it’s impacted what fellowships they are applying to. Some have shied away from Body and Neuro as a result – as people say these are likely the ‘easier’ fields for a machine algorithm to take over..Our seniors are exclusively applying to Breast and IR as these are seen as fields where you ‘need’ a person doing procedures. It seems nuts for me to make such a big decision simply out of fear of the ‘What if” .
That being said, i’m the biggest hypocrite of all – I’ll probably end up doing IR now just so i have a ‘skill set’ that isn’t automatable. I know that means a likely miserable lifestyle and turf wars forever, but hey, until a machine can de-clot a fistula, put in a line, drain some abscess and take some biopsies, ( whatever is left over that vascular, cards, nephrology, etc doesn’t steal in the next 5 years ) at least It’s one less thing for me to worry about.
Do i really believe that diagnostics will get replaced by machines? No. Just being a mid level resident and seeing the complexity in what an average radiologist does day to day makes me think that’s unlikely. It’s easy for a tech person who doesn’t know what the heck they’re talking about to look at us and say “Oh, your job is all about repetition and algorithmic thinking. I can teach a machine to do that.” – But it’s an entirely different thing when so many experienced Attending physicians on this board echo that statement with such bold ness.
After reading post after post after post, fear has crept in. I’m young. I saw this awful radiology job market hit 5 years ago and it’s slow recovery, I don’t think it’s an invincible field by any means. I have 30 years of practice ahead of me. My wife and i are BURIED in loans. We are both working our tails off. I’d like to buy a house one day. Or a car. Or take a freaking vacation that isn’t to a waterpark. I can’t afford for my skills to become worthless in the next 10 years, if that’s what so many here are predicting. This thread has been truly such a bummer for my weekend..
This. As a rising MS4, AI has entered into my psyche more than I would like to admit. I am applying IR/DR (although I really am interested in IR regardless). Out of a class of ~150, probably 8-10 of my classmates were interested in rads until finding about AI. They are going into surgery/IM/derm/etc. It is definitely taking a toll on us med students even if the fear is unfounded and the specialty must be taking a hit from it. Most of us fear we are giving up too much (debt and years of our lives) to risk being replaced by some Silicon Valley overlord. I am still going to do it because I am not sure I could stomach another specialty. -
I’ll watch my words, lest I offend millennials, but it sounds like the younger generation is ruled by its fears. “Silicon Valley Overlord”? [i]Seriously[/i]? I’m far more worried about Washington controlling medicine, although we had a reprieve on January 20th of this year.
NO ONE except for Jimboboy and Zeke Emmanuel have declared our profession dead. Consider the sources. -
Others are declaring us dead too. (From a recent New Yorker article…)
Geoffrey Hinton, a computer scientist at the University of Toronto, speaks less gently about the role that learning machines will play in clinical medicine. Hintonthe great-great-grandson of George Boole, whose Boolean algebra is a keystone of digital computinghas sometimes been called the father of deep learning; its a topic hes worked on since the mid-nineteen-seventies, and many of his students have become principal architects of the field today.
“I think that if you work as a radiologist you are like Wile E. Coyote in the cartoon, Hinton told me. Youre already over the edge of the cliff, but you havent yet looked down. Theres no ground underneath. Deep-learning systems for breast and heart imaging have already been developed commercially. Its just completely obvious that in five years deep learning is going to do better than radiologists, he went on. It might be ten years. I said this at a hospital. It did not go down too well.
-
He doesn’t have a firm enough grip on the full extent of what radiologists actually do, but for some tasks he is right.
The New Yorker article undersells him a bit. He’s one of he main innovators behind the theoretical underpinnings of deep learning and currently leading a lab at Google.
But people on this forum are a bit naive if they think that this is just more of the same and not a legitimate threat.
We should start training fewer radiologists now. So if this scares off a few, that is not a bad thing.
Anyone who is ~10 years away from being a radiologist (ie early medical school) should seriously look into something else.
-
That Hinton article is old news man. He’s a prime example of a non physician talking out of his butt. Need to stop giving platforms to idiots like that. Also, think he backpedaled on those statements later.
-
Calling him an idiot isn’t helping your argument.
I’m a physician and I also think we should start drastically reducing training spots.
A lot of what we do can be generalized as “image captioning”, and that is very much in jeopardy. Don’t kid yourself.
-
I can see academic centers using AI to gradually replace residents in the future, and having attendings fine tuning the interpretation. Then of course eventually attendings will be replaced.
-
Quote from anonirad2012
Others are declaring us dead too. (From a recent New Yorker article…)
Geoffrey Hinton, a computer scientist at the University of Toronto, speaks less gently about the role that learning machines will play in clinical medicine. Hintonthe great-great-grandson of George Boole, whose Boolean algebra is a keystone of digital computinghas sometimes been called the father of deep learning; its a topic hes worked on since the mid-nineteen-seventies, and many of his students have become principal architects of the field today.
“I think that if you work as a radiologist you are like Wile E. Coyote in the cartoon, Hinton told me. Youre already over the edge of the cliff, but you havent yet looked down. Theres no ground underneath. Deep-learning systems for breast and heart imaging have already been developed commercially. Its just completely obvious that in five years deep learning is going to do better than radiologists, he went on. It might be ten years. I said this at a hospital. It did not go down too well.
HIM! HIM!??
He is SO arrogant. People are taken in because he is SO certain of himself. They are also taken in because they don’t know enough to argue against him. I do. He might be right, but chances are very very low. Our real threat, as I have pointed out before is that people who are not sophisticated – say like and MS4, a legislator, administrators, and the like believe him because he is so damn LOUD. (and cocksure)I wrote a letter to the new yorker pointing out that they should probably quote people who have at least some glimmer of an idea of what we do. He is clueless beyond his own mathematics.
Please also read the rest of the article for a more balanced view.
Of course the new yorker didn’t publish it. Maybe because I compared his sophisication to a high school student.
I did, however point out that this was about the 10th rising of “machine interpretation” I had seen in my career, the previous incarnations lie in ashes at my feet.
For the MS4’s out there – don’t run from the fire, run to it. When I got into radiology, it was because I saw the second CT scanner in North America as a medical student, and I thought “I want some of THIS!” Others though were saying it would make radiology so easy that any surgeon, pediatrician, OB, cardiologist, orthopod, etc could do it. BWAH HAH HAH!
We only lost turf when we had too few enthusiastic people to pursue it. Which is what will happen if we have a critical shortage. You need bodies to hold the line.
Others have pointed out that we may go down to AI, but only after every other job in America has gone down – like pediatricians, administrators, FP’s. Leaving only us and theoretical physicists.
-
Quote from Dr.Sardonicus
Others have pointed out that we may go down to AI, but only after every other job in America has gone down – like [b]pediatricians, administrators, FP’s[/b]. Leaving only us and theoretical physicists.
Except that AI is [b][i]not [/i][/b]targeting the bolded. It [i][b]is [/b][/i]targeting radiology.
-
Has anyone actually released any promising data on ai?
Or is this still all big boisterous claims fishing for investments?
-
Quote from DoctorDalai
I’ll watch my words, lest I offend millennials, but it sounds like the younger generation is ruled by its fears. “Silicon Valley Overlord”? [i]Seriously[/i]? I’m far more worried about Washington controlling medicine, although we had a reprieve on January 20th of this year.
NO ONE except for Jimboboy and Zeke Emmanuel have declared our profession dead. Consider the sources.
There actually has been a decent amount of press coverage on AI. Saying silicon valley overlord was tongue in cheek. I was simply reporting the current thinking of medical students. Regardless of whether we are right or wrong, it is understandable that we are weary of entering a career that perhaps has an uncertain future. We are 6-10 years out from being attendings. We do not have houses, millions saved up, we probably won’t make close to the incomes you all do, but we are stupid millennials so what do I know. -
Med students getting scared off won’t lower the number of residents, it’ll just shift the source of where resident labor is acquired.
-
-
-
-
-
“Interesting. The DREAM competitors seem to be primarily from places other than the US. Also, the competition was sponsored by the Laura and John Arnold Foundation… ”
Competition was open to all, and a bunch of the leaders are from US academic departments (not sure what difference that makes anyway).
As for anti MD paranoia, not sure I see it but even if that was their motivation, does it invalidate results of the competition?
The models people create either work or they don’t, and so far it looks like they work.
Something else: it’s very easy to improve on any individual model by combining them all, so final results will probably be near or better than human (just like 2 human readers on average better than one).
-
Dalai,
What does us taking control of AI even mean? Does that mean we as individual radiologists own the technology? Did we invent it? Do we give financial backing? Do they need our permission to use it? The only means of “control” I see is for us to protest, which they will steamroll over for the sake of “greater good.”And even if our eventual role is to supervise the readings, the number of radiologists needed is far fewer than now. And that supervisory role will have an expiry in the shape of a downsloped log curve because the technology will continue to improve.
If you don’t like what i have to say, stop reading my posts. Put me on ignore if there is such an option on auntminnie threads.
-
-
-
-
-
-
-
Quit putting words in my mouth, sonny. You are becoming more and more annoying with each post, and you are sabotaging whatever it is you wish to accomplish, unless that goal is simply to troll, which you are doing well whether you see it or not. Siegel’s answer, which I have posted multiple times, is to get control of the development of the technology. Makes a lot more sense than to sit and quiver in fear over what might be.
I don’t [i]want[/i] to hear anything in particular from you. This is an interesting and valuable discussion, but your constant fear-mongering, shared by one or two others (if that many), does nothing to educate anyone, and is literally scaring medical students away from this field. You “aim to inform so you can prepare”. Very thoughtful. But sonny, you’ve left out quite a few other things for which we should prepare. Nuclear war, civil strife, globalclimateconfusionwarmingchange, zombie apocalypse, Ebola, PMS, asteroid strikes, Takata airbags activating autonomously, Democrats winning back Congress…the list goes on. I suppose you are now cowering in your bed, sucking your thumb in the fetal position.
I’m sorry, sonny, I don’t mean to be this hard on you, but you need a wake-up call. Email Siegel, email IBM. Use your knowledge to HELP our field, rather than this constant fear-mongering. UNLESS [i]THAT[/i] IS YOUR GOAL…
-
Dr. D, Why do you suppose all this AI talk doesn’t include Nucs (sorry, Molecular Imaging)?We were the first modality to go digital, but not word one regarding Unclear Medicine.
I suppose we as techs can do our part in supporting the Rads by continuing to submit barely diagnostic quality exams…malrotated, under/over exposed, incorrectly annotated, clipped bases, poor contrast timing, etc, and watch our computerized overlords display the rotating clock symbol…the Radiologist equivalent of a growl.-
Quote from walshnuc
Dr. D, Why do you suppose all this AI talk doesn’t include Nucs (sorry, Molecular Imaging)?We were the first modality to go digital, but not word one regarding Unclear Medicine.
I suppose we as techs can do our part in supporting the Rads by continuing to submit barely diagnostic quality exams…malrotated, under/over exposed, incorrectly annotated, clipped bases, poor contrast timing, etc, and watch our computerized overlords display the rotating clock symbol…the Radiologist equivalent of a growl.
I see you’ve met some of the techs I’ve known over the years. Remind me to send you the blood-pool images that were submitted as a readable bone scan. The answer to your question might parallel why Macs are rarely targeted by viruses; there are a lot more Windoze boxes out there, and thus there is more bang for the buck. If I were creating AI products, I would concentrate on the larger markets. And no, I don’t think that means imminent doom for anything.-
In terms of this “debate”, while there is “chicken-littlism” always finding the sky is falling, there is also “nihilistic status quoism” — nothing will ever change, things were better the way they were and are, and all these doodads will never really matter. Preparing for the future is important; I am sure PACS was not perceived as disruptive to radiologists, but surely we would have significantly more practicing rads now if we were still on film.
-
Screening exams are the easiest initial targets – lots of data and essentially asking a binary question.
-
-
-
-
-
rads312,
You will be fine even if you go into DR. In fact you may do fantastic for a while. As the realization of AI filters into the general public, there is going to be a CRIPPLING SHORTAGE of radiologists. My feeling is that the job market is going to be gangbusters for the next decade.Once you’re out, pay down your debt as fast as you can. Then build wealth quick-like too.
Dalai,
Am I happy that people will be avoiding radiology? Of course not. IMO, this is the best specialty in medicine. It saddens me to see the way it has become commoditized. I feel that’s what is to come will further devalue our contribution. Still I’d rather people be aware than be caught ignorant. -
am interested – what did he say when he backpedaled?
I don’t think anyone disputes AI will be a disruptor to radiology, probably beginning in 5-10 years. The only question is how.-
@deadwing Well, if AI makes inroads as quickly as some suggest, PDs may wise up and start cutting spots anyway.
Well, those with consciences, anyway.
If anything, this is the time for radiology to be *raising* its standards; not matching students with mediocre Step scores and med school performance. The stronger the cohort, the less easily we’ll be replaced. If radiology becomes the specialty of IMGs and FMGs, it’s all over.
-
If we scare of good applicants and american grads and keep the current positions as is – it’s absolutely going to become an IMG and FMG ruled field.
-
Another incoming MS4 going into radiology. I consider myself competitive for any specialty. I chose radiology because I really love this field and enjoyed my rotation in during third year. However this thread definitely makes me very nervous and concerned. Yes, AI, watson, and all the rumors may not be all true and no one has the power to foresee the future. I try not to be swayed by all these news articles written by MD’s or non-MD’s. Some articles actually seem to show many of the bright sides. What really makes me concerned is some of the posts written by attending physicians in this thread seem to have very gloomy outlook for the future. I was hoping the radiologists are the ones in the driver seat in this rapidly changing field not the passenger seat. This thread makes me think that we have no clue where we are headed and “silicon valley overlords” are the ones who are actually driving the field.
-
Perhaps, but not good to be dishonest or sugarcoat, either.
-
self driving cars and delivery will happen way before radiologists are replaced. AI will take over low paying jobs that lend themselves to easy automation. A server taking your order is easily replaced by a robot or ipads. Companies will use robots to do their work, replacing factory workers. AI will indeed have a big affect, will be a huge revolution and change our lives drastically in the same manner the internet did. After all of this happens, then you can start having the discussion of when will AI replace radiologists…radiologists are extremely protected from being replaced compared to the majority of jobs out there.
-
Quote from striker79
self driving cars and delivery will happen way before radiologists are replaced. AI will take over low paying jobs that lend themselves to easy automation. A server taking your order is easily replaced by a robot or ipads. Companies will use robots to do their work, replacing factory workers. AI will indeed have a big affect, will be a huge revolution and change our lives drastically in the same manner the internet did. After all of this happens, then you can start having the discussion of when will AI replace radiologists…radiologists are extremely protected from being replaced compared to the majority of jobs out there.
So the scary thing is – Tesla’s do drive themselves. And well. And they are testing Uber’s and Semi trucks that do this – and they are doing very well. All my local mcdonalds have automated options for ordering food…But I agree with you 100%. I think there are a LOT more jobs that could be easily replaced before a radiologist’s, such as these I mentioned.
However, even within healthcare – I feel like you could replace a pharmacist very easily with software + a pill counter and dispenser. That hasn’t happened yet. Do we really need a physician to see a patient with URI symptoms for a few weeks to prescribe an antibiotic vs conservative treatment? I don’t know – but I feel like that is a bit easier to make an algorithm for than an MRI of the knee. Who knows.
A lot of what’s said here has discouraged me, some has motivated me..Regardless, I’m just going to do my personal best to be not only well trained, but involved with the legislation and implementation of AI as it continues to creep into our specialty. As so many others have said, we need to get out ahead of it. That being said, If i’ve only got a decade or so that I can count on the kind of income that current Radiologists have, I am most certainly going to plan my finances and investments much more differently.
-
-
-
-
-
-
I have about 10 years out of training and work in PP – I love radiology and find it fascinating and a superb career. However, if I were in medical school right now looking at fields I also think I would have some hesitance to commit to radiology. While the future is incredibly difficult to predict, I do think AI will have a significant impact in radiology in the decades to come, but it is certainly still well off. There are very high barriers to entry in medicine – legal, governmental, ethical etc cause it deals with people’s health and lives. It will require that AI is fairly widespread and accepted in society, meaning it is common place, prior to it being accepted in medicine when dealing with real, live humans.
I don’t think we will see truly widespread AI in society until the advent of quantum computing. That is the huge leap forward which allow machines to develop enough complexity to equal and surpass the human brain. The current binary non-quantum computers are unlikely to push AI into the mainstream of society.
-
I guess [i]no[/i] profession is safe from AI:
[link=http://www.atlasobscura.com/articles/germany-robot-priest-blessu2-religion]http://www.atlasobscura.c…riest-blessu2-religion[/link]-
[link=https://psychcentral.com/lib/computerized-therapy-will-your-next-therapist-be-a-computer/]https://psychcentral.com/…erapist-be-a-computer/[/link]
-
-
Striker said “AI will take over low paying jobs that lend themselves to easy automation. ”
This is a fallacy. Cheap labor will NOT get replaced, because it is CHEAP. Companies replace workers when it makes economic sense, not for the sake of it “lending well to automation”.
Why do you think places like Angola use human labor for mining when heavy machinery can do it much better? Because human labor is cheaper there. Why does Apple use lines of workers in China to assemble iphones when it can create an automated factory? Because it’s cheaper.
We radiologists are expensive. There is greater economic incentive to replace us, even if our task is harder to replicate.
-
Quote from Jimboboy
Striker said “AI will take over low paying jobs that lend themselves to easy automation. ”
This is a fallacy. Cheap labor will NOT get replaced, because it is CHEAP. Companies replace workers when it makes economic sense, not for the sake of it “lending well to automation”.
Why do you think places like Angola use human labor for mining when heavy machinery can do it much better? Because human labor is cheaper there. Why does Apple use lines of workers in China to assemble iphones when it can create an automated factory? Because it’s cheaper.
We radiologists are expensive. There is greater economic incentive to replace us, even if our task is harder to replicate.
Who are the highest paid individuals? CEO’s, sports figures, actors, other entertainers. I haven’t seen anyone attempt to replace them with AI, robots, electric gladiators, or what-have-you. And these folks collectively earn more than all the rads in this country combined.
ALL physician income adds up to 7% of the health-care bill. Ours as radiologists, is what, 1%? Probably less. There are less than 40,000 radiologists in the US. Our collective salary would be something like $10-12B. How much of an investment would it take to replace us? Given the state of technology, more than it’s worth. Keep in mind, many of us do procedures that cannot be done by a machine, or even a P.A. If, Heaven-forbid, a surgeon takes those away, he/she will still have to be paid.
Not every decision is financial, despite what the beancounters would tell you. Those who want the BEST seek out hand-made cars and the like, because human craftsmen can make some things better than any machine. It is the same with health-care.
Jimboboy reminds me of the old saying: “Rarely correct but never in doubt.” We’ve gone from Google being able to recognize the image of a dog to self-driving cars to miners in Angola being used as “proof” of our imminent demise. I think these discussions have become mostly proof of Jimboboy’s imminent psychotic break.-
I dunno. Jimboboy’s posts have definitely strengthened my commitment to contribute to political support for radiologists. The field could use a shudder or two, as long as that energy is channeled to our own good.
-
We should always be wary, but not to the point of paranoia.
BTW, my personal paranoia leads me to say that no politician is trustworthy.-
Dalai,
Funny you should mention CEO’s, entertainers, and athletes. Because those are the jobs that are going to be the most secure (in collective terms, not for individual entertainers). Unfortunately the barrier of entry are exceedingly high.-
Kind of funny how so many people snarl at the salaries of doctors and dentists, but slaver over athletes who make salaries dozens of times higher than ours.
-
Leadership jobs are secure because they have CONTROL. Think about it Dalai, why would they want to replace themselves?
Pro athletes will always have a job because we humans like to gawk at and admire athletic prowess. Human athletic prowess, not robots.
Actors and singers and accompanying entertainment industry are secure because we like to look at attractive people and enjoy creative entertainment. AI will be able to emulate those traits but we as detest knockoffs and fakes. Think about how we treat those entertainers caught lip-syncing, autotune, plagerism, etc.
-
Delai, like Jan alluded, “the human component” can be provided by the clinicians. On that note, what human component have you provided in your dictations today, other than hedging?
As for the patients, they could care less. Heck, significant portion of the population already thinks images are interpreted by machines.
-
Oh boy…
The CEO has control? You’ve clearly never been one, or known one. Ever hear of a Board of Directors? They, and the stockholders they represent, can run the CEO out on a rail if he doesn’t perform.
And we only like to watch humans? Prove it. I submit some would like to watch robots battle. We detest knockoffs and fakes? Why won’t that apply to Radiology? And I (and probably 95% of the rest of radiologists) resent the he!! out of your last comment:
[blockquote] Delai, like Jan alluded, “the human component” can be provided by the clinicians. On that note, what human component have you provided in your dictations today, other than hedging?As for the patients, they could care less. Heck, significant portion of the population already thinks images are interpreted by machines.
[/blockquote] Clearly, YOU buy into the Imaging 3.x hype about how we don’t provide “[i]VALUE[/i]”, whatever that means. Maybe that’s how your practice works but many/most of us try to do better. I think we succeed. And you have absolutely no idea about what patients do or don’t think.
Now that you have resorted to pulling “facts” out of your @$$, you’ve lost whatever credibility you might have had.
Could you possibly [i]try[/i] to discuss the issue without bleating constantly about the imminent death of the field?
-
Oh boy…
The CEO has control? You’ve clearly never been one, or known one. Ever hear of a Board of Directors? They, and the stockholders they represent, can run the CEO out on a rail if he doesn’t perform.
And we only like to watch humans? Prove it. I submit some would like to watch robots battle. We detest knockoffs and fakes? Why won’t that apply to Radiology? And I (and probably 95% of the rest of radiologists) resent the he!! out of your last comment:
[blockquote] Delai, like Jan alluded, “the human component” can be provided by the clinicians. On that note, what human component have you provided in your dictations today, other than hedging?As for the patients, they could care less. Heck, significant portion of the population already thinks images are interpreted by machines.
[/blockquote] Clearly, YOU buy into the Imaging 3.x hype about how we don’t provide “[i]VALUE[/i]”, whatever that means. Maybe that’s how your practice works but many/most of us try to do better. I think we succeed. And you have absolutely no idea about what patients do or don’t think.
Now that you have resorted to pulling “facts” out of your @$$, you’ve lost whatever credibility you might have had.
Could you possibly [i]try[/i] to discuss the issue without bleating constantly about the imminent death of the field?
-
Dude, this jimboy is just trying to push your buttons and get your goat. Of all the threats to radiology, AI is the LEAST credible threat out there.
-
I don’t really know what is the added value of this topic as the linked article doesn’t contain the tiniest bit of novel information. The same people as always once again reiterated their viewpoint – funnily enough this applies to the forum discussion as well. Medical AI is at the peak of the hype cycle now as massive amount of venture capital has been pumped into it. Remaining skeptical at this point is hardly condemnable – we are very likely surrounded by the Theranoses of medical imaging.
Furthermore, we shouldn’t forget the broader context of AI. By the time it will be truly ready to eat out lunch a significant percentage of the workforce will have already been fallen victim of automation… -
Dalai,
BOD in various companies are intertwined and CEO of one company serves as director of another. if you think CEO’s abide by and are subserviant to directors in the real world, you are naive.
I dont think you’re that naive so more likely you’re just being disingenuous.And am I to believe that you think that the same people who call bogus on Lance Armstrong, Ashley Simpson, and Milli Vinilli are gonna hold your CT report to the same standard? If that is true, they are gonna demand that you credits the original authors who discovered the imaging characteristics. You are major B.S.ing if you equate the two.
-
Jimboboy, your posts have degenerated into sheer trash. You offer nothing but your own lofty opinion and want us to bury the field of Radiology based on your trollish preaching.
I wouldn’t think of stopping you, and neither will Aunt Minnie, but your refusal to discuss this issue rationally (and despite your snark, what you have posted recently is far from rational) has done a lot of damage. I really think that’s exactly what you have in mind. And that’s not major B.S.ing. -
Clearly I am making you upset. I suggest you step away.
-
-
-
-
Quote from Jimboboy
Dalai,
Funny you should mention CEO’s, entertainers, and athletes. Because those are the jobs that are going to be the most secure (in collective terms, not for individual entertainers). Unfortunately the barrier of entry are exceedingly high.
You are avoiding the point, which is not surprising.
IF Honda could make a robot football player, the cost of which amortized over time being cheaper than a human player, would that be the end of the NFL? Would a computer that outdid Warren Buffet double the price of Berkshire Hathaway and boot out ol’ Warren? Will GGI replace actors?
[font=”impact,chicago”][size=”7″][u][i][b]NO![/b][/i][/u][/size][/font]
Because there is a HUMAN component to these things, just as there is with Radiology. In your delirium/paranoia/fear-mongering you absolutely refuse to understand this. Most of the rest of us do.
Go find some other children to frighten.
-
-
-
-
-
-
-
DoctorDalai, since radiology essentially exists to give clinicians cues on how to proceed clinically, can you elaborate a bit on the human component of radiology?
-
Welp.
This went to garbage real fast. Come on guys – You are Attendings. You are supposed to lead us. I need a shot of bourbon just to swallow how you treat each other. We are all one the same team here – right?
Maybe your generation isn’t used to getting completely screwed over as mine is ( med school tuition, loan interest percentage, reimbursements, etc ) , but we shouldn’t be fighting one another on this issue – instead, we should be working together to know how to move forward, together.
this bickering isn’t helping.-
My apologies.
-
I like the diverse viewpoints, from DD’s cautious optimism to JB’s deadly serious warnings about the specialty’s future. Take a sprinkle from one and a dash from the other, and we might actually save radiology, from more threats than one.
-
-
You are right, rads312, but to quote my son when he was 2, “He started it!”
I have offered a number of positive actions that could be taken, and I have shared communications I have had with folks smarter than all of us on this topic. If our field is in danger than DO SOMETHING about it; don’t just spread suppositions based on tortuous logic that “prove” our imminent demise. More than half of Jimboboy’s posts are of that genre.
Millennials don’t like conflict (unless they start it) and so the discomfort is not surprising. I’m no fan of bickering either, but given the stakes, as professed by Jimboboy himself, we need to unite on this. But as with most other things in Radiology, we act way to much like Republicans, frittering away our global advantages in favor of individual advantages. (Think self-referral). Can we stop that and talk about some positive actions we could actually undertake that would help the profession?
Or is it more fun to scare the children?
-
“I’m no fan of bickering either, but given the stakes, as professed by Jimboboy himself, we need to unite on this.”
I think the message board has a different view on your claimed aversion to bickering.
As for the need for unity, yes, we need it. But the prerequisite for that is for the parties to agree on the immediacy of the AI issue.
Once that’s done, how do we “control” a develoment we have little leverage over?
-
Your approach is to declare the sky is falling, then run away. At least that’s all we’ve heard out of you.
Reread the line you quoted and your followup. This is typical of your word-twisting. Yes, I’m “bickering” with your fear-mongering, and I will continue to do so. You are causing our field immediate damage without the slightest shred of trying to do any good for anyone.-
-) chuckle chuckle(-
And your approach is to pretend it won’t happen while closing your eyes and covering your ears.
You want proof of the urgency of AI matter in radiology?
But before I go there, what proof do you have other than ill-informed opinion of yourself and another?
I told you this topic will be getting more coverage 2 years ago. It has.
I will present the actual proof on a platter within 1-2 years in the form of an actual AI generated report !
(for clarification-cuz you seem to need a lot of that- I will not be working on the AI myself and presenting it to you. What I will probably show you will be an article or a paper done by others. …Because I’m a practicing radiologist, not an IT or research person)
I will enjoy rubbing that pie on your face. (more clarification, not a real literal blueberry tart pie, but a figurative one. Metaphorical pie)
-
Jimboboy, you continue to put your own instability on display, and I thank you for doing so on behalf of radiologists everywhere. It’s sad, really, as you have posted some very helpful answers on other threads, but this topic has apparently driven you mad.
As far as “Ill-informed opinon of (my)self and another”…I do believe you are the one posting your own navel-gazings as absolute, irrefutable facts, as you have done here. You are saying that you [i]will[/i] [size=”5”][i][b]EVENTUALLY[/b][/i][/size] show us the “proof on a platter” in the form of an AI generated report in 1-2 years. Which tells us that such things do not exist as yet. (I’m sure you would have found one if you could.) And my response will be….So what? EVEN IF AI can [size=”5″][i][b]EVENTUALLY[/b][/i][/size] do what you fear (which is a YUGGGGGE IF), that does not automatically imply that it [i]will[/i] be adopted as our replacement. What I have advocated, as described by Eliot Siegel, far wiser than all of us put together, is that we utilize the technology [i]ourselves[/i] as it evolves, that we get in on the development, rather than sit in our own verbal excrement and p1$$ and moan about the falling sky.
I’ll ask you for the last time…just [i]what[/i] do suggest we [i]do[/i] about your imminent disaster? The only thing you’ve offered is that we should learn to do something else. How incredibly helpful.
And you are going to “enjoy rubbing pie in (my) face”? What a petty little man you are. Based on your very unusual syntax, I am [i]still[/i] convinced you are the disruptive poster that has plagued AuntMinnie for many years in many different guises.-
You keep asking me for a solution on how we can control AI in radiology. I keep telling you that there may not be a good answer to that. What part of that don’t you understand?
-
Also, proof is here already, but you have to let yourself see it. It’s been here for 2 years.
Google AI can decipher photos and distinguish people.
That in itself should be enough. The reason why you can’t see it is because of various personal bias. Like self importance, it being not applied specifically to radiology, etc.
-
Quote from Jimboboy
Also, proof is here already, but you have to let yourself see it. It’s been here for 2 years.
Google AI can decipher photos and distinguish people.
That in itself should be enough. The reason why you can’t see it is because of various personal bias. Like self importance, it being not applied specifically to radiology, etc.
That is NOT proof at all. Already discussed and dismissed. You’re starting to panic.-
I think things will change, however quickly or slow, but I foresee a significant amount of radiology training be dedicated to infomatica and computer science training.
I feel that 50 years from now, I may lament that my trainee cannot read a CT as well as me like how my attending lament that I don’t read plainfilm as well, but they will be able to achieve better diagnostic accuracy by working WITH the AI, not unlike how I can do quite well by using cross sectional imaging.
The key here is that we need to work with the AI and take charge of the DATA.
Fortunately, a lot of people have already take notice. Our local private practice have been approached with and refused to providing data to AI companies. Those data will worth a lot.
We will worth a lot for awhile because the machine will need to learn from us.
Maybe one day the youngsters will lament how we sell off the field by profiting of the interpretation data we generate while allowing AI to advance like how we see the guys who sold their practice to the coporate devils.
But it’s something we have to do. Like some of those boomers, we will have no choice.
-
Sorry Dalai, one more reponse.
In regards to my pettiness for my planned relishing of metaphorically rubbing pie in your face. Yes, it is petty. But I am susceptible to human foibles like anybody else. You have not been particularly cordial in our discussions. in fact you have been relentless in your belligerence. So I feel like you are in need of some symbolic ridicule.
-
OK, I’ll be cordial.
I wish to formally, respectfully request that Dr. Jimboboy grace us with information based in reality. Also, if it wouldn’t be an inconvenience, it would be simply lovely if Dr. Jimboboy could be so kind as to share his solution of just what it is we who have not been blessed with the knowledge and insight so clearly radiating from his scintillating persona are supposed to do in regard to his dire predictions.
By the way, I find it fascinating in the extreme that Dr. Jimboboy, who expects us to hearken to his IT expertise and subsequent prophecy, forgot his old ID, which was used for all of six posts, and was unable to resurrect said ID even though the tools are quite clearly available upon sign-in. -
Seriously, how many times do I have to tell you? There may be no viable way for radiology profession to protect our turf. Go ask Siegel how to do it.
I do wonder though, maybe we can all decide to walk out when the time approaches. As professionals tho, it’s dicey.
As for the previous ID, the email address I was using became defunct. Recovery attempt was not worth my time after that.
-
Too many feels on this board.
I’ve been following the progress of these new techniques pretty closely, and I would say that the technology to replace most of diagnostic radiology will be there in a decade.
The real question is how quickly things will be implemented, approved and integrated into healthcare delivery.
So people starting their careers now will be safe for a while, but if I were an MS1 I would not consider diagnostic radiology or pathology viable specialty choices.
-
-
And in regards to your paranoia that all the people who irritates you are one person.
Full disclosure: I had only one other ID in the past: Magnifier.
I posted under that name back in 2012 when I posted about Google AI seeing Cats.
[link=http://www.auntminnie.com/forum/tm.aspx?m=350978&high=google+cats]http://www.auntminnie.com…50978&high=google+cats[/link]
Unfortunately I forgot my login. No other ID. So relax
-
-
-
I read 90% of this thread, but have no idea what your real disagreement about AI is.
Would you both please in a nutshell state your opinions about AI?
Not opinions about each other or what the other has posted (we all got that), but about AI.
For example,
Dalai, do you believe that in the next 2-10 years AI will have the same impact (minimal) as CAD did in the last 2-10 years? I think so, but I am not sure.
Jim, do you believe that in the next 2-10 years AI will replace (>50%) of radiologists? I think so, but I am not sure.
My opinion is that it will have a significant effect in the next 2-10 years. The business will change, but not go away. It’s actually a very exciting time for Radiology. If utilization remains constant, a few Radiology jobs will go away. I would be wary of going to medical school today and getting a bottom 25% residency spot. Those in the top 50% will have exciting busy careers.
-
Quote from NYPhD
If utilization remains constant, a few Radiology jobs will go away.
This is a very pertinent observation that reminds me of this recent article written by a practicing radiologist who also happens to be a medical AI researcher. In this he coined and elaborated rather novel term by “radiology escape velocity”, pointing out that automation has to be staggeringly fast in order to have any effect on the radiology workforce.
[link=https://lukeoakdenrayner.wordpress.com/2017/05/08/the-end-of-human-doctors-radiology-escape-velocity/]https://lukeoakdenrayner….ology-escape-velocity/[/link]
-
-
-
-
-
-
-
-
If there’s nothing we can do, why are you constantly vomiting your fear-mongering all over these pages? So you can [i][b]EVENTUALLY[/b][/i] “rub pie in people’s faces” on the [i]very[/i] slim chance that you are correct? Just WHAT is it you want radiologists to do? I keep asking you that…what part of that don’t [size=”4″][b][i]YOU[/i][/b][/size] understand? Seriously, Jimboboy, or Provastin, or whomever you really are…You’ve been at this for two years, doing nothing but telling us about imminent doom, with absolutely no solution whatsoever. None. Bupkiss.
Why do I somehow have the feeling that IF “eventually” ever looks like it might possibly come about, you’ll be attempting to sell us some “solution” that will be our “one and only” salvation. [size=”4″][i][b][size=”2″][/size][/b][/i][/size] -
Your only “answer” is that there’s no answer. So you come here and wallow in your fear-mongering and think you’re doing the field a favor by letting us know “The End Is Near”???
We should all decide to walk? So we FORCE the powers that be, whoever that is, to use whatever AI is available? Brilliant. I’m just overwhelmed by your analysis.
Whatever else you are, Dr. Jimboboy, you are far from stupid. So put your intellectual prowess toward finding a solution, rather than just giving up. I have a very strong suspicion you have some solution up your bu…I mean sleeve, which you will miraculously reveal to us, the great unwashed, when you think we are softened up enough to beg you for it.
But do continue to scare the children if you wish. I’ll follow with a note to medical students reminding them that you have absolutely NOTHING in the way of proof of your day-mares.-
I don’t have to fear monger anything. Rest of the world is running with it. And if you’re gonna accuse me of starting it, you give me too much credit.
And for the last time, the proof that we are appoaching AI medical image interpretaton is : Google AI can interpret photos now. It does not take a rocket scientist to see the connection. You certainly do not need actual AI produced report to see it.
And again (ad nauseum) there may be nothing you and I can do to prevent AI takeover. All it requires is a thought experiment: What woud happen if someone made a machine that can interpret medical images as well as us? Answer: powers-that-be will replace us.
What if the machine is marginally worse? Same thing because it will be cheaper.
-
[link=https://www.nature.com/nature/journal/v542/n7639/full/nature21056.html]https://www.nature.com/na…/full/nature21056.html[/link]
I would think that would be enough proof that these techniques are something to be taken seriously.
And what was done in this Nature paper is really basic – it could easily be improved.
Radiology is a little protected by the ways our images are different from natural images (grayscale, sometimes 3D, high resolution), but not much.
Anyone in the middle or end of their career is fine, but I don’t know why there is such hostility to people suggesting that current Med students need to worry about this, because they need to be aware of it.
-
Here we go again. Google can recognize a cat in a 2D photo, so our lives are ruined. And the ability to recognize the margins of a 2D skin lesion is not that much different. This is not proof of anything. I just saw on America’s Got Talent last night a chicken that played the piano. Should concert pianists worry?
Jimboboy, you have been repeating the same stuff with the same tenuous evidence again and again and again and again and again. AND IT GOES NOWHERE. You have the demeanor of a terrier tugging on a rope…you just won’t stop with the incessant fear-mongering.
Your rather limited, repetitive argument, lacking much logic, goes something like this:
[ul][*]Google can recognize a cat on a jpeg image.[*]Therefore by next week AI will be able to interpret 3D images of complex lesions in the body and write its own report that won’t hedge like Dr. Dalai does. (No proof offered of any of these contentions)[*]AI will [i]of course[/i] be cheaper (again no proof offered) so we will therefore all be summarily dumped from our high-paying jobs. [/ul]
Did I miss anything? Aren’t you getting tired of living in fear, or does it help to spread it around? -
Here’s how I think about AI: Quoting the great Hale Irwin – “I never underestimate my opponents, but I never underestimate what I am capable of”. Take away…we should take AI very seriously, but keep in mind there’s likely a synergistic outcome that we contribute to. Radiologists will never become obsolete, but I do believe AI will have a similar impact as PACS. We will become more efficient over time, thus demand for rads will decrease slowly over time, but there will always be demand.
Now, shoot me down all you want, but let’s be nice! 😉-
I think you are precisely on target, Rad-ical. No shooting from here. You’ve said what I’ve been saying, but more eloquently.
-
I think people on this board don’t have enough of an appreciation for how much has changed since 2012 (and took a bit longer to accelerate progress).
For those not familiar with XKCD, it’s a great comic made by someone with a lot of math and computer science experience.
[link]https://m.xkcd.com/1425/[/link]
And an explainer here:
[link=https://www.explainxkcd.com/wiki/index.php/1425:_Tasks]https://www.explainxkcd.c…/index.php/1425:_Tasks[/link]
At the time this comic was written (~2014), the idea of getting a computer program to recognize if a picture was of a bird was considered impossibly hard.
The alt text says, “In the 60s, Marvin Minsky assigned a couple of undergrads to spend the summer programming a computer to use a camera to identify objects in a scene. He figured they’d have the problem solved by the end of the summer. Half a century later, we’re still working on it.”
And here’s the key point – this comic is out of date. Minsky’s problem is essentially solved, although still some room for improvement.
[link]https://arxiv.org/abs/1703.06870[/link]
And all the technique requires is a labeled dataset.
We like to think that all of our schooling and special knowledge protects us, but it doesn’t really. The things that were hard for computers were the things we did without thinking thanks to billions of years of evolution.
“Moravec’s paradox is the discovery by artificial intelligence and robotics researchers that, contrary to traditional assumptions, high-level reasoning requires very little computation, but low-level sensorimotor skills require enormous computational resources.”
Essentially it’s easier for a computer to solve differential equations than tell a joke.
And although we may think it’s simple to distinguish between a dog and a cat, historically for computers this had been very hard. But that is no longer true as of the past year or two.
And I think there’s an argument that perceptually, that’s harder than many of our diagnoses. I know that will be controversial to many because we can distinguish between dogs and cats without thinking (as can almost anyone), but try describing how you distinguish between the two. It’s a lot easier (for me at least) to describe a PE.
What we do is much harder for other humans than distinguishing between dogs and cats, but that is no guarantee it will be for a computer.
-
Quote from anonirad2012
And all the technique requires is a labeled dataset.
Which is an acknowledged bottleneck though, labelled medical imaging datasets are hard to come by and are much smaller than necessary. Furthermore the tiniest deviation from the ground truth can cause further systematic errors.
-
That is not correct Balint, neural net is precisely different because it does not compound errors.
And dataset is exactly what IBM acquisition of MERGE in 2015 was about. It acquired the dataset to feed Watson.
Thanks anonirad. I’m glad you joined the discussion.
-
Quote from Jimboboy
That is not correct Balint, neural net is precisely different because it does not compound errors.
And dataset is exactly what IBM acquisition of MERGE in 2015 was about. It acquired the dataset to feed Watson.
Thanks anonirad. I’m glad you joined the discussion.
That depends on a lot of things e.g. whether its supervised or unsupervised learning, or whether the training dataset is labelled. AFAIK IBM’s dataset is not labelled thus its value is limited.-
Agree with Jimboboy in post #128. If AI says “high likelihood of pancreatic cancer, would recommend tissue biopsy for diagnosis”, that could, quite conceivably, be sufficient for further intervention in the future. It’s not as though radiologists are infallible as it is.
-
@Phil Shaffer
There are worlds of difference between jobs changing – e.g., the way in which virtually all cataract surgeries are now performed by making small incisions vs the old way of basically opening up the whole cornea and then having to suture it back 270° – and the question of a job *disappearing*. If AI performs DR, what are trained DRs supposed to do? Go back to residency, I guess.
-
What Phil said. Spot on. It’s a far more eloquent version of what I’ve been saying. Harness and utilize the new paradigms, and don’t sit and quiver in fear that they will take you over.
ZZ, I’m quite disappointed in your rant, which basically has as many logical and “debating” fallacies as you’ve accused me of displaying. You Millennials have a hard time with conflict you didn’t start. And to blame me for Jimbobob’s keyboard diarrhea is simply foolish on your part. Go to your room and think about what you’ve done.
While awaiting your forensic analysis, I did my own. Here are some of JimboHEGEMONRADbobdMarkupMDboy’s posts containing dated predictions:
[link=http://www.auntminnie.com/forum/fb.ashx?m=462948]http://www.auntminnie.com/forum/fb.ashx?m=462948[/link]
[link=http://www.auntminnie.com/forum/fb.ashx?m=462950]http://www.auntminnie.com/forum/fb.ashx?m=462950[/link]
[link=http://www.auntminnie.com/forum/fb.ashx?m=463036]http://www.auntminnie.com/forum/fb.ashx?m=463036[/link]
[link=http://www.auntminnie.com/forum/fb.ashx?m=463227]http://www.auntminnie.com/forum/fb.ashx?m=463227[/link]
[link=http://www.auntminnie.com/forum/fb.ashx?m=473396]http://www.auntminnie.com/forum/fb.ashx?m=473396[/link]
[link=http://www.auntminnie.com/forum/fb.ashx?m=515654]http://www.auntminnie.com/forum/fb.ashx?m=515654[/link]
Notice that in the past month or so (not linked) those predictions have been pushed further out, rather than made more imminent. Interesting.
Many of you weren’t here in the era of the troll-posters HEGEMONRAD and dMarkupMD and Chopra. I was, and I can tell you Jimboboy’s endless repetition of the same stuff with slight variation, and glancing, word-twisting swipes (lying) about other’s questioning of his veracity is IDENTICAL to those entries. Go back and read them yourselves.
In some of those cases, I got similarly caught up in defending the field from these doom-sayers. I stand by every word I wrote then, and every word I write now. AuntMinnie moderators have seen fit to close the accounts of these trolls one by one. You’ll note “Bence” has disappeared; JimboHEGEMONdMarkupBoy got a little careless with that one. Sadly, I’m sure the main character will be allowed to continue to ply his foolishness, and that’s fine, but I’m counting on the rest of the community to make sure the kids looking at Radiology are not scared away by this madness.
My bet stands as I wrote it. I don’t let trolls gerrymander the rules to rig things. Clearly, it won’t be accepted, which comes as no surprise to anyone.
In the meantime, please reread my 2016 RSNA AI post:
[link=http://doctordalai.blogspot.com/2016/12/artificial-intelligence-at-rsna-im.html]http://doctordalai.blogsp…igence-at-rsna-im.html[/link]
I think this is the far more realistic description of the path that will be followed. AI will become a game-changing TOOL for us. Like PACS. Like CT. Like everything else we utilize for patient care. But it needs to be [i]our[/i] tool.
-
Jan,
In support of my thoughts on the “black box” issue.Clinical labs today are essentially a black box. Blood goes in, the printer spits out that serum potassium is low.
The clinician then adds a bag of potassium, without questioning the result most of the time. They may repeat occassionally to confirm or if there is concern for hemolysis. But they do not ask for supporting evidence of why potassium was interpreted as low.
same thing with RF, CEA, AFP, PSA. These markers may not be perfect tests but (+)test automatically lead to greater scrutiny for disease. (-) test automatically leads to a lot of skepticism.
-
*yawn*
Anybody seen Man in the High Castle? Good show.
-
Indeed it was. I wonder if they’ll do a third season
-
watch this “Life in Pieces”. Not often am I laughing so hard I can’t breath.
netflix.
-
-
-
-
-
-
-
This seems like the most likely outcome to me. Despite having an autopilot, human pilots still fly the plane, at least for critical portions.
I hear a lot about the uninvolved aloof radiologist who simply cranks out reports, but in my practice, and in a lot of friends’ practices, we field dozens of calls a day from referring doctors with questions about reports. Is the patient improving, was the finding there three CT’s ago, stuff like that. Hard to replace that with a machine in the foreseeable future.
Also, what about suboptimal studies? A good chunk of what we look at in community practice is suboptimal, whether for motion, poor contrast bolus, suboptimal positioning….any machine learning system better be prepared to deal with those issues. Will they be able to decide whether to read a study vs. requiring it to be repeated? All issues that need to be worked out prior to our being replaced.
And any of these systems will have to be FDA approved. My gut is that we’ll become more efficient, and slowly need fewer radiologists. Probably in 30 years be employed computer jockeys troubleshooting 500 CTs a day. But not obsolete.-
Uh. shouldn’t I feel compelled to inform y’all if we’re hosed?
Anyway, I think most radiologists are coming around to the realization. Not on my account per se. But I think it good nonetheless.
And yes, radiologists will have supervisory roll in the beginning. For that, fewer number of diagnostic radiologists will be needed. But as machines get better and better, that number will fall farther.
Eventually who will educate the supervising radiologists? The machines? The question then becomes: why have human diagnostic radiologists at all?
-
Dalai, you said it will never happen in our lifetime. You even said not within our kids lifetime. Anyway, thanks for coming around.
-
Quote from Jimboboy
Dalai, you said it will never happen in our lifetime. You even said not within our kids lifetime. Anyway, thanks for coming around.
Love your Alinsky tactics. I have not “come around” and everyone here sees your manipulation of other people’s statements. Keep it up and you’ll end up in government. As a Democrat. And that is NOT a complement.
I still don’t see [b]it[/b] happening in my lifetime, and probably not yours. [b]IT[/b] being machines completely replacing us, that is.
We are ever so lucky to have a great seer such as yourself to make sure we all can prepare for the end. Which no one but YOU is predicting, by the way, not even Anon.
-
-
My understanding is that a major problem with neural networking/AI is that it is a “black box.” It cranks out an answer & it is not true intelligence – It doesn’t know what its looking at – just based on a statistical model “this image contains cancer with 78% certainty” for example.
AI can not explain how it comes to a conclusion. This is a problem when findings are subtle, when images are used to guide therapy or track response to treatment.
Since radiologists most frequently make perceptual errors, this promises to reduce missed findings on screening exams or incidentals but I don’t think it will actually kill radiology for the MANY different tasks we are act to do when “only” interpreting images (set aside all of the other things radiologists do).
Correct me if Im wrong; Im not a AI researcher. -
My understanding is that a major problem with neural networking/AI is that it is a “black box.” It cranks out an answer & it is not true intelligence – It doesn’t know what its looking at – just based on a statistical model “this image contains cancer with 78% certainty” for example.
AI can not explain how it comes to a conclusion. This is a problem when findings are subtle, when images are used to guide therapy or track response to treatment.
Since radiologists most frequently make perceptual errors, this promises to reduce missed findings on screening exams or incidentals but I don’t think it will actually kill radiology for the MANY different tasks we are act to do when “only” interpreting images (set aside all of the other things radiologists do).
Correct me if Im wrong; Im not a AI researcher.-
That is a good question. But that question is an artifact of how we practice medicine right now. In the current model, the clinician questions how we came to a conclusion based on the findings. With that supporting evidence, they frel safe to carry out the treatment.
But why does medicine have to follow that model in an AI future? Why wouldn’t the clinician blindly trust the AI and carry out the treatment without the need for evidence, simply because AI has called it correctly before? They may not care that it’s a black box as long as it has a decent track record
-
The black box paradigm suffices as a check against perceptual errors (which we are prone to). Due to the black box, it fails to improve on cognitive errors (which we are not as prone to). Thus, radiologist+AI is better for all of our patients and hugely valuable to society and our profession.
As an example:
Imagine a scenario where AI has called a high likelihood of cancer on a screening mammogram, and thus correctly assigned BR0.
The physician (radiologist, or in our dark fantasy: surgeon or FM med doc) working the patient up on diagnostic day/future needs to know why this is called or a few different problems arise:
(1) where is the cancer is located and what are its characteristics for dedicated follow-up imaging (do i need mag views, spot compression, ultrasound, etc).
(2) on follow up imaging, what is the extent of the tumor for surgical planning
(3) What are the suspicious features for rad-path correlation
(4) on follow-up studies, do the new suspicious features match the previous tumor characteristics.
I think knowing that cancer is present is critical, but isn’t the only important question to guide management. Guiding management and changing clinical decision making is the value of our profession; AI will greatly assist by reducing missed findings. -
I should add I am not an AI researcher either.
But that was a great question. I hope my answer makes sense.
On that note, the tone of the message board has become a lot more constructive than before (notwithstanding certain you-know-who) on this topic. It tended to degenerate into “not in our lifetime”, “who are they gonna sue”, “Terminator/2001 Hal references” before.
I thank you for taking this seriously, which it deserves
-
I fail to see why AI cannot address each if those.
-
Quote from Jimboboy
I should add I am not an AI researcher either.
But that was a great question. I hope my answer makes sense.
On that note, the tone of the message board has become a lot more constructive than before (notwithstanding certain you-know-who) on this topic. It tended to degenerate into “not in our lifetime”, “who are they gonna sue”, “Terminator/2001 Hal references” before.
I thank you for taking this seriously, which it deserves
Not an AI researcher. So I’m curious why you feel the timeline laid out by Erickson and Siegel is too conservative. You’ve really only given minimal supporting evidence for your position, and spent a whole lot of time making sweeping statements that people are wrong (you “fail to see why AI can’t address each of those”, which is not really a rebuttal). And when I asked why you brought it up and kept at it, you said you felt compelled to. Which isn’t an answer either.
Bravo, 7/10 for trolling.
-
Why do you guys consider only full replacement as a doomsday scenario?
How about a network that is extremely sensitive, but not too specific. Get a radiologist to filter false positives (also, easier to implement than full replacement). Make such duo read 100 CTs a day (“Agree” and “Dismiss” buttons under slides with pertinent images). Get convenient and convincing performance/outcomes data and show that you can replace 5 human radiologists with this hellish bundle. Sell this idea to policy makers/admins. Now you need 5x fewer radiologists, and you can leverage those who are left into working for 100K. Boy, if I were an admin, wouldn’t it be nice to crush these basement-dwellers into the life of despair!
My angle of attack as an evil admin would be – sell AI as a “support tool”, gain the required evidence for the bureaucrats and then through decreased demand make these pesky doctors with 7 years of post-graduate education belly dance for me. Dance, doctah, dance!
P.S.
As far as the state of the art for deep learning is concerned, it is improving weekly (literally). Take a look at what individual groups do in public competitions [link=https://grand-challenge.org/All_Challenges/]https://grand-challenge.org/All_Challenges/[/link] -
I’ve had enough. There’s only one way to put a stop to this nonsense, and that is to make the players put up or shut up.
So, Dr. Jimboboy, here’s the duel, I mean deal. Put your money where your keyboard is. It’s time for a friendly little wager, something that will let us see how much you believe in your endless drivel.
First, we must define the test, sort of a radiological Turing test if you will. I declare the endpoint of your mad predictions as some AI interpreting [i]all CT, MRI, U/S, NM/PET, mammographic, and CR/DR examinations fed to it[/i], generating a report that is indistinguishable from a human-generated report, and being placed in a production environment where it [i]replaces[/i] a human radiologist. Someone must get fired and replaced by the AI to fulfill the prophecy. Your worst fears come true.
IF this occurs within 2017, I will donate $1,000 to the LEGITIMATE charity of your choice, as long as it does not support terrorists. Or you. If your nightmare scenario does not occur, YOU will donate $100 (yes, I’m being nice) to MY favorite charity, RAD-AID International.
IF this occurs before the end of 2018, I will donate $900 and if it doesn’t, you will donate $200. And so on. I’ll even go so far as to freeze my level at $500 for the five years after 2022. If I’m alive, which I hope to be. After that, we’ll renegotiate. Anyway, your stake will keep going up by $100/year. But you’ll still have a job, so big deal.
There are many on this board who know me personally, and know I mean what I say and that I will indeed live up to this promise (and will hold me to it). This is for real, and for a good cause, Rad-Aid, which will most certainly clean up.
So put up or shut up. -
[link=https://pubs.rsna.org/doi/abs/10.1148/radiol.2018184007?journalCode=radiology]https://pubs.rsna.org/doi…?journalCode=radiology[/link]
Read the fourth paragraph.
Quote from DoctorDalai
I’ve had enough. There’s only one way to put a stop to this nonsense, and that is to make the players put up or shut up.
So, Dr. Jimboboy, here’s the duel, I mean deal. Put your money where your keyboard is. It’s time for a friendly little wager, something that will let us see how much you believe in your endless drivel.
First, we must define the test, sort of a radiological Turing test if you will. I declare the endpoint of your mad predictions as some AI interpreting [i]all CT, MRI, U/S, NM/PET, mammographic, and CR/DR examinations fed to it[/i], generating a report that is indistinguishable from a human-generated report, and being placed in a production environment where it [i]replaces[/i] a human radiologist. Someone must get fired and replaced by the AI to fulfill the prophecy. Your worst fears come true.
IF this occurs within 2017, I will donate $1,000 to the LEGITIMATE charity of your choice, as long as it does not support terrorists. Or you. If your nightmare scenario does not occur, YOU will donate $100 (yes, I’m being nice) to MY favorite charity, RAD-AID International.
IF this occurs before the end of 2018, I will donate $900 and if it doesn’t, you will donate $200. And so on. I’ll even go so far as to freeze my level at $500 for the five years after 2022. If I’m alive, which I hope to be. After that, we’ll renegotiate. Anyway, your stake will keep going up by $100/year. But you’ll still have a job, so big deal.
There are many on this board who know me personally, and know I mean what I say and that I will indeed live up to this promise (and will hold me to it). This is for real, and for a good cause, Rad-Aid, which will most certainly clean up.
So put up or shut up.
-
52% is piss poor specificity for detecting an occluded vessel on cta
-
I cant believe this topic is coming back up
-
Have to agree with kpack. Even I tire of this topic.
But I’ve maintained that AI generated report will come by 2020. Radiologist displacement start by 2025.
So I found it odd that Dalai (who countered that it will never happen in our professional life) decided to place the deadline in 2017/2018.
But what are you saying? did I win?
-
-
-
-
-
-
-
-
-
-
Curious like Dr. Dalai. What’s the motivation for posting if you think we’re hosed?
If it were me, and i thought radiology had no future, I wouldn’t waste my time posting. -
Anon, I appreciate your attempt to provide reasonable discussion here. It is a breath of fresh air in this rather stuffy room.
I’m going to need your help. Here is your quote from Cornell:
[blockquote] We present a conceptually simple, flexible, and general framework for object instance segmentation. Our approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation mask for each instance. The method, called Mask R-CNN, extends Faster R-CNN by adding a branch for predicting an object mask in parallel with the existing branch for bounding box recognition. Mask R-CNN is simple to train and adds only a small overhead to Faster R-CNN, running at 5 fps. Moreover, Mask R-CNN is easy to generalize to other tasks, e.g., allowing us to estimate human poses in the same framework. We show top results in all three tracks of the COCO suite of challenges, including instance segmentation, bounding-box object detection, and person keypoint detection. Without tricks, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners. We hope our simple and effective approach will serve as a solid baseline and help ease future research in instance-level recognition. Code will be made available.
[/blockquote]
It seems to me to be a huge reach from this object detection to medical diagnosis. Now to be upfront, I have seen the Watson demos, which probably use a somewhat similar approach, and they do appear to recognize pathology. Ironically, the standard demo detects a PE. We have talked about this before.
[i]Given enough time[/i], I have no doubt that Watson or his descendants could “learn” to interpret medical imaging studies. I also have no doubt that computers can learn to emulate the finest diagnostician either, or any other field that requires sensory input but no manual manipulation. And I wouldn’t be surprised, [i]given enough time[/i], that the descendants of the DaVinci robots mate with Watson’s kids and produce an AI surgical robot that can do a better job than humans. [i]GIVEN ENOUGH TIME. [/i]
No one has said ignore AI, no one has said it will NEVER be able to do what we do. Those statements are simply Jimboboy’s effort to put words in people’s mouths who don’t wish to freak out along with him. Because NEVER is a very long time, isn’t it?
I refuse to panic, and I refuse to incite others to panic. I will go out on a limb: AI is a [i]tool[/i], nothing more, nothing less. IT WILL NOT REPLACE US. There is nothing in development with that in mind, and even if there was, the legislative hurdles would be huge. And if this is all predicated on money, well, someone tell me how much Watson costs.
Now until someone shows me something that can outdo Watson’s primitive performance, this discussion will remain nothing more than a lot of hot air and innuendo.
And I repeat, and this was discussed at more than one talk at RSNA…Google recognizing a cat does NOT imply the capability to interpret complex imaging, any more than the first fish to walk out on land was able to build a spaceship. The first step? Maybe. That’s about all.
