Chinese AI Beats 15 Doctors In Tumor Diagnosis Competition (thenextweb.com) 79
An artificial intelligence system called BioMind has managed to defeated a team comprised of 15 of China's top doctors by a margin of two to one. The Next Web reports the details: When diagnosing brain tumors, BioMind was correct 87 percent of the time, compared to 66 percent by the medical professionals. The AI also only took 15 minutes to diagnose the 225 cases, while doctors took 30. In regards to predicting brain hematoma expansion, BioMind was victorious again, as it was correct in 83 percent of cases, with humans managing only 63 percent. Researchers trained the AI by feeding it thousands upon thousands of images from Beijing Tiantan Hospital's archives. This has made it as good at diagnosing neurological diseases as senior doctors, as it has a 90 percent accuracy rate. Further reading available via Xinhua.
Impressive (Score:5, Interesting)
Re: (Score:1)
Al ain't no Chinese name I ever heard of.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
"Al ain't no Chinese name I ever heard of."
You need to come out from under your bridge sometimes.
https://en.wikipedia.org/wiki/... [wikipedia.org]
Re: (Score:2)
Re: (Score:1)
So the doctors diagnosed 15 cases per minute (225/15)? Amazing that they were 63% accurate when they were only spending 4 seconds per case! How much time would they spend per case if they weren't competing against AI?
Re: (Score:2)
Strange indeed. The Xinhua article never mentions how long the doctors took.
If it were true, it could be that the 15 doctors handled separate cases. That would mean 225 cases/15 doctors=15 cases/doctor and 30 minutes/15 cases = 2 minutes/case.
Re: (Score:1)
It's like those "Turing test competitions" where average people can't tell the difference between a human and a crappy chatbot. If you read the actual dialogues, you'll find that the human did a good job pretending to be a computer, the conversations w
Re: (Score:2)
"This is really amazing. It is like computers are good at image recognition. I see a lot of potential in this AI."
Indeed. It will replace all those doctors pretty soon.
Better still, and I can't wait, its cousin will replace all those lawyers, who did nothing but 'read the book'.
Re: (Score:3)
This is really amazing. It is like computers are good at image recognition. I see a lot of potential in this AI.
Actually medical diagnosis was the first challenge AI succeed at...back in the 1970's. It wasn't until the late 90's that AI succeed at another major feat (beating the world champ at Chess). There really isn't anything surprising or impressive about this feat.
The problem with AI systems in medicine are the doctors who don't want to use the technology. Step into a hospital sometime and look at their computer systems which will often be older than what you can see in the computer museum. The second prob
Re: (Score:2)
Re: (Score:2)
To a different test with different information.
Re: (Score:3)
A probable tumor in a scan is often followed up with a biopsy or exploratory surgery for confirmation.
Re: (Score:2)
More details would be nice (Score:5, Insightful)
Did the AI give any false positives?
Did the doctors correctly diagnose any cases that the AI did not?
Whilst this is great news, I hope doctors use it as a learning and aid tool instead of a full diagnosis suite without a review by the specialist.
Re: (Score:1)
I would imagine that your question should be asked. Considering the area i.e. human health - this maybe he case although the majority approach to any also automatic authority is such that this probably be implemented the way the HR system in UK was done and the overwrite function will not be present.
Re: (Score:3)
The big problem with AI at the moment is that we don't understand how it makes decisions. You can ask a doctor why they made a certain diagnosis, but you can't ask an AI in many cases. So at best the AI result can prompt a human doctor to look again, but it can't give much in the way of hints as to why it disagrees with the human.
This is a huge problem and one which the EU has addressed with the GDPR, which gives you the right to know how decisions were made and on what grounds. That prevents companies simp
Re: (Score:2)
Thanks GDPR, I can continue getting my hand-made artisanal - but explainable - diagnoses which are _wrong_ instead of getting inexplicable _yet accurate_ diagnoses from software.
Re: (Score:2)
Did the doctors give any false positives?
Did the AI correctly diagnose any cases that the doctors did not?
Whilst this is great news, I hope AI uses the input of specialists as a learning and aid tool instead of a full diagnosis suite.
AI as an invaluable support tool for doctors (Score:5, Interesting)
Machine learning will not soon replace human doctors, I think, but you can see it becoming a powerful support tool for doctors, hopefully finding ailments earlier as well. Fascinating.
Re: (Score:2)
you can see it becoming a powerful support tool for doctors
As long as the tail doesn't wag the dog there, I agree.
I'm sorry Dr. Bill [youtu.be], you cost 1 Million dollars a year to pay, while McBox Boxieface over there costs $500/month for power plus maintenance. And can literally be in two places at once. Why don't you go back to school and choose a better degree this time?
Re: (Score:2)
Dr. Bill collecting a million dollars a year is not a net positive for society if Boxie McBoxFace really can contribute the same services
Re: (Score:2)
i had read previously of Watson helping doctors in diagnosing a baffling case of Leukemia that seemed totally untreatable, Watson correctly indicated that the patient had 2 kinds of Leukemia at once, which the human doctors didnâ(TM)t realise.
With two kinds of leukemia at once, you're surely a goner anyhow, so the actual value might not have been that great for the patient.
Still, knowledge is knowledge and a value in itself.
Re: (Score:2)
If you want to differentiate, it's more reasonable to differentiate between "strong AI" and "weak AI." Strong AI is what most people mean by AI, whereas "weak AI" is "cool stuff that's not really intelligent."
BioMind (Score:2)
I like the name, since all minds are biological except this one.
ahnuld says (Score:1)
it's not a tumah!
Ways to think about machine learning (Score:1)
Yeah, but... (Score:1)
my cat beat those same Chinese doctors on the same test [/sarc]
Let me know when that AI beats doctors from Johns Hopkins, or CTCA, or City of Hope, etc.
It's not that I, as a tech freak, doubt that AI can become better at such tasks, but rather that I have worked with Chinese guys with PhDs who in reality were little better than high school graduates in the west were back in the 80's (not now perhaps given the American educational system plunge). I simply doubt the use of Chinese doctors as a yardstick. Oh,
Beats 15 CHINESE Doctors! (Score:1)
At least 10 of them didn't even go to medical school - they just got some papers printed up.
Humans... (Score:3)
...need not apply.
In other words... (Score:2, Flamebait)
Celebrate if and only if you wish more bad diagnosis in China.
In other tests, doctors failed 100% (Score:2)
When doctors stare at a sample of blood they were consistently unable to identify that the sample had sickle-cells.
Just saying "AI did it" doesn't actually make this anything more that a centuries-long list of "a medical test has been invented". It's good and all that but there seems to be some idea that "AI" means "voodoo".
Re: (Score:1)
A medical test was done but we don't know how it did it. That's the AI part. No one told it how.
Or do you somehow mean the program itself invented a whole new medical test? Because that would be even stronger AI.
failure rate (Score:2)
what i really find worrying is that doctors were only able to detect 66% of tumors, so a lot of people get false diagnostics for something very deadly.
So... (Score:1)
Chinese doctors: 66% success
Flip of a coin: 50% success
Chinese doctors are a bit closer to the coin flipping rather than to AI performance. Maybe that, rather praising AI, we should instead blame the professional capacities of the doctors. Furthermore, consider that since diagnosis is mostly based on images, AI can (and probably does) perform some basic image processing that can greatly enhance the detection of subtle features. To my experience doctors usually work with raw images (and
False negatives. (Score:1)
False negatives are really important here.
100 cancer cases. 75 cancer diagnosis. 25 non cancer diagnosis. 75% success. 25% dead due to misdiagnosis.
100 non cancer cases. 75 non cancer diagnosis. 25 cancer diagnosis. 75% success. 0% dead due to misdiagnosis.
The way the AI deals with this can make or break it.
No license! (Score:2)
The self-governing monopoly trade union known as the AMA will see to it that this software never enters the public domain. The beach houses and Ferraris of featherbedded radiologists should never come under threat.