AI Isn't Replacing Radiologists (worksinprogress.news) 42
Despite AI models outperforming radiologists on benchmark tests since 2017, demand for human radiologists has reached record highs. American diagnostic radiology residency programs offered 1,208 positions this year, up 4% from 2024, while average salaries hit $520,000 -- 48% higher than 2015. Over 700 FDA-cleared radiology AI models exist, yet only 48% of radiologists use AI at all. Models trained on standardized datasets lose up to 20% points accuracy when deployed in different hospitals. Radiologists spend just 36% of their time interpreting images, with the majority devoted to patient communication, teaching, and administrative tasks that current AI cannot perform.
No surprises here (Score:2)
Re: (Score:2)
AI just picks up on simple patterns and the problem was that you would feed a bunch of data into an AI for radiology and it would take the easiest route to an answer which would often be catastrophically wrong.
That's the problem when you are right 90% of the time you're basically killing 10% of the people who needed care and treatment.
To be fair this is before the current models but at least five or six years ago one of the radiology
Just another tool (Score:1)
Re: (Score:1)
The number 1 reason for unnecessary deaths
Solution... Easily accessible transmission X-ray full body scanners in malls and workplaces would allow people to pass through once a month and let an
Re: (Score:2)
Solution... Easily accessible transmission X-ray full body scanners in malls and workplaces would allow people to pass through once a month and let an AI look for anomalies that when detected would be passed to a reviewer and then signal the health app on people's phones.
That would cause cancers from the x-rays, because the radiation from it is cumulative. That's why we don't use them to pick your shoe size anymore. Yeah, we actually did that for years.
Re: (Score:2)
If you build a machine which detects and treats cancer noninvasively, then if the machine causes cancer and treats it before it's a problem, I don't see the issue.
It's basically a machine that can put the eggs back in their shells.
Re: (Score:2)
Umm... Yeh, and cutting people open to perform surgey causes holes in bodies.
Snide remarks do not improve your argument.
If you build a machine which detects and treats cancer noninvasively, then if the machine causes cancer and treats it before it's a problem, I don't see the issue.
The reason you don't see an issue is because you don't understand the actual risks involved. When my wife had cancer, treatment of which did not require radiotherapy, she was warned about said risks from the multiple CT scans she would be getting for the next 5 years. Been a while so I don't remember exactly what the doctor said but here's Wiki's quote on it:
An abdominal or chest CT would be the equivalent to 2–3 years of background radiation to the whole
Re: (Score:2)
An abdominal or chest CT would be the equivalent to 2–3 years of background radiation to the whole body, or 4–5 years to the abdomen or chest, increasing the lifetime cancer risk between 1 per 1,000 to 1 per 10,000.[134] This is compared to the roughly 40% chance of a US citizen developing cancer during their lifetime.
Did you catch that last line? A _single_ body CT will give you about the same chance of getting cancer as the average American has over their lifetime...
From X-ray [wikipedia.org].
I'm afraid you've completely misunderstood that, possibly construing "compared to" as "equal to".
The quotation actually says two things: (1) Each US citizen has about a 40% chance of developing cancer at some point in their lifetime; and (2) an abdominal or chest CT scan increases the risk BY roughly 1 in 1000, or 0.1% (from 40% to 40.1%). Also note that it says the CT scan is about equal to 2 to 3 years of background radiation. If cancer risk increased by 40% every 2-3 years, everybody would
Re: (Score:2)
No, there's no radiologist conspiracy against AI, any more than there's an Uber drive conspiracy against Waymo cars. In both cases, the technology is still immature and limited. Both technologies are making headway, but there's a long way to go.
Title of job referring to a small portion of job (Score:5, Insightful)
Radiologists spend just 36% of their time interpreting images
Think this translates to a lot of the 'AI should replace them' positions people think about.
I'm a "software developer". There are certainly some code-heavy times, but there are days where I don't 'code' at all, and days like today where I've only messed with around 6 lines of code. A minority of my job is taken up with tasks that are even hypothetically in the discussion for AI replacement. It just so happens the LLMs tend to suck at most of my niche as well, but even if it were spot on for prose-to-code it still would only reduce a smaller fraction of my job than an outsider would guess....
Re: (Score:1)
The radiologist having a tool to analyze images is like the programmer having an IDE that checks for syntax errors and provides auto complete. It will help them to spot the stupid mistakes and enabled them to put their concentration on the logic instead of searching for minutes where the semicolon is missing. And as the IDE never gets tired reminding you of the missing semicolon, the image AI never gets tired of highlighting relevant parts of an image. Humans on the other hand have their ups and downs.
Re: (Score:2)
the IDE never gets tired reminding you of the missing semicolon, the image AI never gets tired of highlighting relevant parts of a... colon?
Wait... what? (Score:5, Interesting)
"Only" 48% of radiologists use AI? That actually seems to indicate AI has made huge inroads into the field and is well-accepted.
And radiologists spend "just" 36% of their time interpreting images? That's a pretty hefty chuck of their time!
This article - BTW WTF is worksinprogress.news? - seems to be AI generated.
Re: (Score:2)
AI has "slowly" been gaining adoption in radiology fields for a while now; machine learning algorithms have been being used in image analysis since the early 2010s. Most of these are the traditional place for AI, improving a skilled worker's productivity, and with machine learning image analysis systems that has been trained for some time, the AI can spot relativel
Re: (Score:2)
BTW WTF is worksinprogress.news? - seems to be AI generated.
A printed magazine that publishes 6 issues per year since August 2020.
Radiologists spend time on patient communication? (Score:3)
Radiologists? Patient communication? Come on, now, pull the other one. Next to forensic pathologists, radiologists are the specialty least likely to communicate with patients. Maybe they're the absolute least likely, though the communications would be rather one way with the forensic pathologists. They do write reports, but they're short and typically intended for other doctors anyway.
Re: Radiologists spend time on patient communicati (Score:2, Insightful)
Re: (Score:2, Troll)
Re: (Score:3)
A lot is hanging on what's meant by "patient communication".
I worked for almost 10 years with pathologists, and there is a whole lot that goes into communication. A gastroenterologist friend once said to me "pathologists are the only ones in a hospital who make an actual diagnosis". This means that when a pathologist or radiologist is writing their report they are laying down some key markers for how a patient will be managed as much or more than strictly documenting a diagnostic finding. That might mean c
What if AI goes out of control? (Score:1)
Re: (Score:2, Flamebait)
What if the moon crashes into the earth or a super villain puts lasers on sharks? We need to be prepared!
Re: (Score:1)
And AI has an actual intelligence and combined with robotics decides to take over and subjugate humanity and we become it's slaves?
Rather than debate the likelihood, I'll just address this question as asked.
What would be the difference? Most of us are already slaves, just given the illusion of options. Which master do you wish to serve is hardly the question asked of free people.
On top of that, it's a very short leap from your propose scenario to, "No need to enslave humans when robotics can do all they do, and more than likely reach efficiency metrics created by the AIs specifically so that robotics can meet them better than humans."
I'm not trusting a fucking programmer (Score:1, Flamebait)
JFC look at all the CVEs out there. Programmers are garbage, and you want their software to make medical decisions? No thanks.
Re: (Score:2)
And those are just the security flaws. But isn't that all software?
Re: (Score:2)
And that's why I live in the woods and keep The Computer in one disused bedroom and only go online when I'm cranky and want to piss all over what's making the world a worse place this week. Offline is life.
Re: (Score:2)
I hope you keep a shotgun next to the computer, just in case
Re: (Score:2)
Offline is life - I'm taking that one
Re: (Score:2)
Re: (Score:2)
> Why do all people think the purpose of AI would be to replace a human instead of augmenting it?
Because, when you augment humans, you either get more work done, or you make some employees unnecessary. Tractors just augment humans also, but when you have them, you suddenly need less people on the field. And while there might be currently some waiting time to the hospitals, there is also upper limit to how much healthcare people need. Once you have augmented people enough to hit that limit, you can improv
Re: (Score:2)
AI will just allow us to get by with fewer radiologist and if they make 500K a year, that's good savings.
Re: (Score:2)
AI caused this problem (Score:3, Interesting)
I already commented about this [slashdot.org] on another story, but AI caused this problem back in 2015-2017 when it was all over the news that AI would begin reading studies. This caused enough of a percentage of med students to specialize in other areas, leading to the big shortage we have now.
As far as AI being used, it is very subtle and minimal. For example our hospital uses iCAD for mammogram imagery [icadmed.com], and all it does spit out a marker file that then can be overlaid on top of the images the Radiologists read, showing areas where it identified extra density and the like. These are just little marker icons showing the radiologist an area that they should review in more detail. Very basic, but it does help keep some things from slipping through.
Anyway AI created this problem, and now we're almost to the point we need AI to fix it. But really, like every single other thing in healthcare, it comes down to liability, and who gets sued when a screw-up happens and who is paying the malpractice insurance that covers it. Right now the doctors bear that responsibility, so they are in demand and make the big dollars. AI reading images would decrease their demand and thus their salaries...
Re: (Score:1)
Obviously and clearly false. To end up with fewer diagnostic radiologists, you need fewer (or unfilled) residency positions in diagnostic radiology. While interventional radiology historically fills all its positions on match day, diagnostic radiology does indeed have some number of unfilled positions on that day, but not ones that persist into the academic year, due to the (continued) high desirability of the residency. Broadl
Re: (Score:2)
We're not talking about 2025, but the last decade. What's your source and what was the trend over that period of time?
Re: (Score:2)
https://pmc.ncbi.nlm.nih.gov/a... [nih.gov]
A 2022 study found that half of medical students who consider specializing in radiology as 1 of their top 3 choices are concerned about the impact of artificial intelligence (AI) on the field
https://www.auntminnie.com/ima... [auntminnie.com]
If results from a survey of medical students in the U.S. are any indication, the misperception that artificial intelligence (AI) will replace radiologists poses by far a bigger threat to the specialty than the technology itself, according to research published online June 27 in Current Problems in Diagnostic Radiology.
After surveying over 150 medical students from radiology interest groups at medical schools around the country, a research team led by Dr. Christian Park of Penn State Health Milton S. Hershey Medical Center in Hershey, PA, found that nearly half of the respondents were less enthusiastic about radiology due to AI.
"The significance of this is not to be understated in that half of potential candidates to the specialty feel as though there is limited opportunity due to an emerging technology such as AI," the authors wrote. "These sentiments have the potential to create downstream effects such as reduction in recruitment to the field of radiology or even medicine as whole."
https://www.sciencedirect.com/... [sciencedirect.com]
A total of 156 students responded with representation from each year of medical school. Over 75% agreed that AI would have a significant role in the future of medicine. Most (66%) agreed that diagnostic radiology would be the specialty most greatly affected
Nearly half (44%) reported that AI made them less enthusiastic about radiology.
On and on.
Everyone's Favorite Radiology-Bot (Score:2)
Here I wrote what could be everyone favorite radiology-bot:
function interpretImage() {
echo "Appears benign."
}
Re: (Score:2)
you forgot a line.
function interpretImage() {
echo "Appears benign."
echo "Follow up again in 12-18 months."
}
Yet! (Score:2)
TFTFY.
AI may never be as clever as a skilled radiologist (Score:1)
I've heard variations on this many times:
My wife went in for x-rays of her gall bladder, pain and such. The radiologist referred her to a kidney specialist, he happened to see something on a kidney that he thought ought to be looked at further.
Sure enough, tumor. Kidney removed. Potential tragedy averted.
I've heard other examples, shoulder x-rays disclosing lung cancer for example.
This is perhaps not easy to train an AI to. More to the point, I want a skilled and inquisitive radiologist giving my x-rays a