Eine Revolution in der Notaufnahme: KI-Modell übertrifft Ärzte bei Diagnose und Behandlungsentscheidung
https://www.ynet.co.il/health/article/syzb5nqczl#autoplay
Eine Revolution in der Notaufnahme: KI-Modell übertrifft Ärzte bei Diagnose und Behandlungsentscheidung
https://www.ynet.co.il/health/article/syzb5nqczl#autoplay
13 Kommentare
Submission Statement:
# A new study published in the important journal Science reveals that OpenAI’s artificial intelligence model correctly diagnosed 67% of cases in triage – compared to 50%-55% among experienced doctors. When looking at who plans better treatment after diagnosis, the gap deepens in favor of AI: „You can easily imagine a system running in the background on the electronic record and detecting diagnostic errors.“
As a woman, I think I would be more inclined to trust ai over a male doctor. The amount of women that have died because their health concerns were brushed off is so shocking. So many stories of women presenting with symptoms multiple times only to be brushed off or told just to lose weight only to wind up having stage 4 cancer or something is so depressing. Something that also stuck with me was a redditor sharing a story about how their mothers health was ignored and after she passed apparently there was sooo much tissue from her endometriosis that a doctor had said she would have been in unimaginable pain almost constantly. And anyone who has had that tissue bind to their bowels or other parts of their body would very much understand how painful it is.
That’s interesting. I wonder if you have different ai and it might overlap and increase detection rates
I’ve had a lot of health issues over the years. Cancer survivor to boot. There are certainly good diagnosticians out there, but they seem few and far between. I’ve been a patient of every major NYC medical system – some of the best doctors in the world in theory – and have been completely shocked.
ERs seem to be the worst. Really horrible place to need to go in the worst moment of your life.
I’m afraid of death, but I’m afraid of spending the last hours of my life in an ER even more.
If AI can improve this, I’m all for it. I am required to get yearly colonoscopies and a few years ago they started using AI to identify polyps early.
This is obvious to anyone deep into Ai. Most of medical diagnosis is perfect for Ai. With more wearable tech we will have less and less people relying on overpaid doctors. Many doctors were google searching prior to Ai.
There was a study that claimed that AI was better at reading radiology than experienced radiologists. Turns out you didn’t even need to provide it the x rays, the metadata was enough for it to outperform.
A bit strange, not needing to see the X-ray to analyze it.
One should be careful about believing any miracle AI gains.
ETA: the AI in this paper was reading the output of a clinician, it did not identify relevant factors from the start.
The article points out the study only worked as all the information needed was already written out, which is not how real life works, until AI can look at an x-ray and see what is correct or wrong then it can not do the job of a doctor
Did anyone read the study? This statement is hype. The study did not show AI doing real triage, treating patients, or safely catching diagnostic errors in live clinical workflow. It tested text-based cases. The EHR claim is speculation dressed up as a finding, so the submission statement is basically useless as a summary.
I’m sceptical. Most diagnoses are relatively straight forward, which is why we have endless protocols in the ER. For those, AI should perform better because doctors in the ER in the UK go from new SHOs / CT1s to experienced Consultants.
However, doctors should be good at the fringe diagnoses, where they may not get the actual diagnosis but recognise a problem that sits outside a protocol, develop a differential diagnosis, and refer to specialists who may then pin the diagnosis down more accurately.
The difficulties in medicine – and why it is hard – is spotting the 20% who just deviate from the normal pattern enough to warrant caution and further investigation. For example, ST elevation MI, or aortic dissection? Pulmonary oedema or bronchospasm? Septic shock or cytokine storm?
Misleading article headline. As other said it had all things written out and did guess correctly what was the issue but that is not how life and diagnosis works. Good support tool but nothing else.
I am looking forward to that Star Trek future where if you’re ill you can lay on the med bed get hooked up to all the sensors and have the computer monitor you and use the vast knowledge of medical knowledge also from previous patients outcomes to provide doctors with the recommended treatments and pick-up on them rare occasions where it’s not what they think it is
Man, the last thing I want to know when I hobble into the ER, is that the MD will get the diagnosis wrong half the time, and a 67% score by an AI is a huge improvement.
That’s impressive, but also a bit scary if you think about real-world impact.
AI doing better in controlled tests is one thing, but ER situations are messy and unpredictable.
Feels like it should assist doctors rather than replace them anytime soon.