Diagnostic software triples rate of full recovery in stroke patients
Specialised AI-assisted tool used on 60,000 people works up to an hour faster than traditional scans, allowing almost half to make total return
We can cut the number of trainee doctors and pay them less then.
Excellent.
This is one of the interesting issues we get with AI across many areas and not just medical. It eliminates the trainee entry level jobs but the more skilled and senior people are still needed to set direction, monitor the results and flag errors. As they leave or retire they wont have understudies to replace them so unless the AI has advanced enough by them to replace them too we will have a problem.
I doubt it’s “AI”
It’s more likely a classifier trained on MRI scans. It doesn’t use an LLM, just some sort of neural network. You train the network by showing it as many MRI scans as you can get, and tell it whether each is from a stroke patient or not.
These things can be as accurate as humans, and aren’t thought of as AI.
Some doctors are remarkably bad diagnosticians. Chap I knew in France was treated for months for shingles. No improvement. It was scabies all along.
One GP failed to diagnose my slipped disc. A secretary at work diagnosed it. “How do you know?” “My uncle walks just like that.”
Tbf my current GP wants a face-to-face appointment with me whenever possible, not a phone call or exchange of e-mails. It’s because, I infer, she believes she can learn a lot just by looking at me. “Your colour is so much better than last time and you look less scrawny.”
I suspect she weighs me partly to see how well I can balance on the scales. She also emerges from her office to watch me walking.
@Andyf ,
Really a non-problem.
If you train AI *correctly* in medical issues and analysis, you train it to “dissmiss” all the results that are definitely not what you’re looking for, and only pass on the rest for evaluation by human eye.
You *vastly* decrease the amount of results that need to be checked by eye, giving junior doctors a much higher “quality” of results to learn from, in much smaller volume.
And they’ll still have to check with their seniors anyway on any unclear result where the AI *did* flag.
So overall you reduce the amount of labour, while making the labour itself more valuable. In theory freeing up doctors to do actual doctoring.
Unless of course, you talk about the NHS, where any reduction in work is a prime excuse for an extra round of golf, or *shudder* padel, or…
And with the way modern labs work… wouldn’t even impact junior doctors.. Since the actual prescreening the AI would do is now done by a trained technician in the lab itself, well before any doctor, student, trainee, junior, or otherwise ever gets the lab results passed on.
For exactly the same reason stated above: elimination of proper negative results, reducing necessary eyeball-hours in the higher tiers.
dearieme:
That’s certainly possible. It’s also possible that F2F is billable while phone calls and emails aren’t.
Strokes need treating asap to achieve the best outcomes. The trouble is that bleeds into the brain have very similar effects to a blood clot, but the required treatments are opposite. So a system that cuts the diagnosis time by an hour is to be welcomed, pity that the ambulance now takes two hours to arrive.
“It’s also possible that F2F is billable while phone calls and emails aren’t.”
It’s the NHS; none of it is billable. GPs are paid mainly on how many patients are registered at their surgery. They get paid a bit more for politically correct doctoring – diagnosing diabetes, giving Covid jabs, prescribing statins, and so on. How much the latter can add up to per annum I don’t know. Anyone? 10% of income? Less?
Many moons ago my GP suspected all was not well in my bowel region. Had a scan done. The radiologist concluded I had well developed bowel cancer and urgent surgery was required. A surgical specialist agreed with this interpretation of the scan results as did my long-experienced GP.
So I ended up under the knife, my bowels were substnatially removed and inspected at whixh point the sugeon muttered “Shit”! It seems I had an impacted appendix that all three medics had failed to recognize and certainly did not require the extensive surgery I had just undergone.
I emerged from the anesthetics to the sight of my surgeon standing at the end of my bed, in blood stained gown muttering, “Sorry! sorry!”. Of course, at that point, I had no idea what he was talking about!
Wonder how an AI would have interpreted the original scan.
I’m sceptical. Of two American AI diagnosis startups, one consistently underperforms humans, and the other turned out to be vaporware. Both have been financial flops. The job of a press release is to make exaggerated claims, but only triple? C’mon, you can do better than that!
I’m not convinced AI is ready for important tasks. I asked ChatGPT which English team has won the most top tier titles:
Manchester United 20 (Last title: 2012–13 (Premier League))
Liverpool 19 (Last title: 2019–20 (Premier League))
If it can’t get something like that right, what chance has it got with important stuff? (Yes, I know, Bill Shankly, “…a matter of life and death…” 🙂 but still.)
“I emerged from the anesthetics to the sight of my surgeon standing at the end of my bed, in blood stained gown muttering, “Sorry! sorry!”.” At least your surgeon, of whichever sex, behaved like a man.
My last op was unfinished when I came to: the bugger was still guddling in my chest. Neither the surgeon nor the gasman came round afterwards to apologise.