back to article So this is how that terrifying killer AI will end us... by pushing us down hospital wait lists?

Artificially intelligent software can help slash the wait times for patients anxious to see the results of their X-ray scans, it is claimed. Long waiting times are the bane of hospitals worldwide. And, so, researchers from the King’s College London and the University of Warwick in old Blighty believe they can alleviate the X- …

  1. JassMan Silver badge

    This system won't reduce staff

    Unless they produce a system which can do a full diagnosis, it won't save any time. Most of the delay in radiography is due to the time to write up the diagnostic report.

    I am shocked that any radiography takes 30 days. By that time any ailment will either have cured itself or gone from urgent to critical.

  2. Hegelworm Messerchmitt

    Worse Case 1% - Isn't That 3000 Fails Per Month?

    https://www.youtube.com/watch?v=xDgkmXAsvL8

    The training set is obviously way too small (according to the article, it's only ingested what equates to a month's worth of data). Another big failing point of AI tends to be the lack of context (i.e. metadata), yet it shouldn't take much to cross-reference this with the patient's medical history, age, gender etc.

    1. Richard 12 Silver badge

      Re: Worse Case 1% - Isn't That 3000 Fails Per Month?

      1% erroneous "normal" when it's "critical" is truly horrifically bad.

      Even the worst-case, knackered after a week of 16-hour shifts, radiologist is better than that given five seconds of glancing at the film while wheeling the patient back out.

      1. Baldrickk Silver badge

        Re: Worse Case 1% - Isn't That 3000 Fails Per Month?

        As someone who spent a week with his arm in a temporary cast with a "broken wrist" only to have it taken off when turning up for the proper cast to be done with a "yeah, that's not a break, that's just a growth line", I can tell you that they still make mistakes.

        1. MonkeyCee Silver badge

          Re: Worse Case 1% - Isn't That 3000 Fails Per Month?

          "I can tell you that they still make mistakes."

          Not sure that's a mistake. It's common practise to treat all possible wrist fractures as worst case situations. Since the potential cost is pretty huge, both in loss of function and potential court case as a result, but the cost is fairly minimal. Being in plaster has minimal side effects :)

          I broke my wrist, and was in plaster for four months. The bones I'd definitely broken should have been fine after six weeks, but because of a potential scaphoid fracture I got an extra six weeks in plaster. It's inconvenient, but I'd rather put up with shaving left handed for a couple of months than risk not being able to grip anything with my right hand. In the same way that I expect you'd prefer to spend a week in plaster and not have a broken wrist, than have a broken wrist and be told it's fine.

      2. Ian Michael Gumby Silver badge

        @Richard... Re: Worse Case 1% - Isn't That 3000 Fails Per Month?

        Radiologists working 16 hour shifts?

        Not in the US. Must be the NHS to blame.

        Seriously... If this is the future of socialized medicine... count me out.

        (And yes, I know many Radiologists. )

    2. W.S.Gosset Bronze badge

      Re: Worse Case 1% - Isn't That 3000 Fails Per Month?

      > The training set is obviously way too small (according to the article, it's only ingested what equates to a month's worth of data)

      Actually, it was trained on 12+yrs data, but from "only" 3 hospitals: "January 2005 to May 2017 at our institution, a publicly funded university hospital network consisting of three hospitals"

      So yes, you have geographical/localcultural bias, but pretty good otherwise in terms of dataset design/acquisition given budget constraints.

  3. Anonymous Coward
    Anonymous Coward

    less effective at correctly labeling critical and non-urgent scans

    never mind those few "unfortunate" cases (sorry about your loss, can I help you with anything else?), look at the broader picture - SAVINGS! :/

  4. Rupert Fiennes Bronze badge

    Diagnostics is not the problem

    In my experience, NHS scanning and X Ray scans are completely siloed in that they don't distribute electronic copies of anything: you get a nice printed scan. Even consultants have issues getting electronic copies.

    1. Anonymous Coward
      Anonymous Coward

      Re: Diagnostics is not the problem

      'In my experience, NHS scanning and X Ray scans are completely siloed in that they don't distribute electronic copies of anything..'

      After a bit of industrial grade 'unpleasantness' with one of my hands a while back, I can at least confirm that at my local NHS hospital the X-Rays are all electronic...go see consultant, get sent down one corridor to get the X-Ray of hand done and by the time I got back to the consultant's room he was looking at the thing on his PC to see how well his repair job was working out. (Bugger wouldn't give me a copy though...).

  5. Anonymous Coward
    Anonymous Coward

    Intriguing

    I also had this problem.

    Simple fix: when you go for the scan, take a brand new el cheapo pendrive and use the oh-so-helpful machine with a card slot in it intended for folks wanting in utero pictures.

    Then when the staff ask what the actual heck you are doing (tm) mention its for a research project in quantum neural networks.

    Note: if said medical staff go very quiet and display a pair of BSOD messages in both eyes, don't blame me!

    Note 2: only works for scans of the: head, eyes, spinal cord etc.

  6. W.S.Gosset Bronze badge
    Thumb Down

    Misclassification- actually 24-52%

    I just had a quick glance at the research, but noted this:

    > In terms of misclassifications, of the 545 radiographs classified as normal by our AI system, five (1%) had critical and 95 (17%) had urgent findings detailed within the reports. On rereview of these five critical radiographs, the AI interpretation of normal was unanimously believed to be correct in four instances. Similarly, for the 95 urgent radiographs, 36 (38%) were normal on rereview.

    The good news is, the AI was RIGHT and the humans initially WRONG for 4 of the 5 'critical' misclassifications. Not so good: the AI was wrong for 60% of the 'urgent' cases misclassified as 'normal'.

    Me, I'm liking that humans-in-a-hurry tended to be over-cautious: false-positive rather than false-negative.

    But if you want to really spook yourself re the AI's non-usefulness, have a look at the FULL misclassification data in Table 3.

    24-52% misclassification rates. Meep.

    1. a_yank_lurker Silver badge

      Re: Misclassification- actually 24-52%

      For an accurate diagnosis, one needs to understand nuances that might not show well in a x-ray but the clinical symptoms indicate. Some kidney stones show up poorly in a x-ray, if at all. But you have a patient with kidney stones. A competent person will understand the problem and will make allowances for it. Artificial Idiocy not likely unless someone programmed it to take into account other clinical symptoms. But your typical program is not a clinician.

      1. W.S.Gosset Bronze badge

        Re: Misclassification- actually 24-52%

        Yes, if you look at Table 2 you'll see the results of their separate/preceding NLP work on interpreting the clinician's own notes, and it has much much better results.

  7. spold Bronze badge

    Gets smarter..

    Assume it will quickly learn which patients are unlikely to survive and lower their appointment priority as there is no point in upping it .

    This would improve care to the treatable patients obviously.... no problem with that is there....

    (alternatively "You have an appointment at the morgue department at 10am next thursday, please bring a sheet").

    1. VikiAi Silver badge
      Unhappy

      Re: Gets smarter..

      Well, technically human medical staff have already been doing that since before recorded history. It's called triage.

      It sucks (for the medical staff who have to make the calls, too!) but without infinite resources it is inevitable.

  8. Haefen

    Treating symptoms instead of disease.

    "Long waiting times are the bane of hospitals worldwide."

    That is the disease. Treat that and such symptoms will disappear.

    That disease is easy to find, it even suggests where we should start looking: " particularly in publicly-funded healthcare systems".

    Elsewhere in society when people rush and kill others we hold them accountable.

    Why give publicly-funded healthcare systems a double oh license? Because they are in a rush to save money for special government projects?

    Remove that freedom from accountability and many of these symptoms will disappear.

    These issues are greater in Canada which has outlawed competition in many areas of healthcare. If a Canadian wants timely effective treatment they have to leave the country, risk underground options or even illegal treatment. Which is not all bad, it reminds those Canadians what it was like to get an abortion decades ago.

  9. Anonymous Coward
    Anonymous Coward

    It could of helped my father

    My father's artificial hip broke at the top, so he had to have extensive surgery to remove the stub that was left in and then have a replacement fitted. All went well and he was discharged. He continued to complain about pain but this was dismissed as normal due to the nature of the operation. However three weeks later his leg broke, it transpired he had been discharged from hospital with a fractured leg. The fracture had been caused during his operation, but as the x rays were digital and the area of the operation was at the top of the leg and the fracture was halfway down his thigh bone. So everyone who looked at them zoomed straight in on the top and missed the fracture further down...

    1. Version 1.0 Silver badge

      Re: It could of helped my father

      A conversation with a trained doctor would have found that very quickly - a conversation with an assistant might (or might not) have progressed to the point where a doctor found the problem but if AI decides it's not a problem then you are done for.

      Machines are here to help doctors diagnose a problem - but these days we just want to get rid of the medical professionals and rely on Artificial Idiots. If a doctor gets it wrong then they get blamed and everyone learns, when AI get's it wrong we just say that statistically they are right most of the time.

  10. Paul Hovnanian Silver badge
    Trollface

    For profit, private medical care

    What's a waiting list?

  11. Alister Silver badge

    “It is no longer feasible for many Radiology departments with their current staffing level to report all acquired plain radiographs in a timely manner"

    Maybe they could employ some more staff?

    1. MonkeyCee Silver badge

      Easy answers are wrong :)

      "Maybe they could employ some more staff?"

      What a great idea. It's not like there is a shortage of radiologists. Certainly not worldwide, and definitely not in the UK....

      Here's the 2017 report:

      https://www.rcr.ac.uk/system/files/publication/field_publication_files/bfcr185_cr_census_2017.pdf

      Shortfall is roughly a 1000 consultant level positions.

      The money spent on outsourcing imaging would cover ~1300 consultants. So the budget exists, but no staff.

      ~10% of clinical radiologist posts are not currently filled. 70% of them haven't been filled for more than a year.

      Of the clinical radiologist's trained in the UK each year (~300), 83% join the NHS.

      ******

      So it seems that there is money to pay for them, a training program that the majority of graduates enter the NHS and yet we're still short. Vacancies are going unfilled for years. So "hire more bodies" doesn't work when there aren't any more.

      Overseas recruitment has dried up too. There's plenty of demand all over the world, and the UK is not currently a great place to relocate too.

      Note, this is nothing new. This has been coming for at least a couple of decades. Workload increase of 20-30% pa, hiring increase of 2-4% pa.

      1. VikiAi Silver badge

        Re: Easy answers are wrong :)

        Maybe they could make radiology a more appealing career to follow?

        Where is this supply/demand pricing thing the hard-nosed capitalists keep insisting is the panacea to all the world's woes?

    2. Medical Cynic

      Maybe they could employ some more staff?

      Yes. Only 5 years or so AFTER basic medical qualification to train as a radiologist.

      You'd have to fund more posts and then wait for the manpower to qualify.

      Better would be to improve working conditions etc so that existing radiologists don't retire early, emigrate or go part time. All due to stress and relatively poor remuneration in the NHS.

      Many radiologists are also going part time and/or stopping extra sessions [eg waiting list initiatives] due to punitive pension taxation rules.

  12. Anonymous Coward
    Anonymous Coward

    Heh it's Warwick

    Anon for obvious reasons but Warwick have systemic problems in the entire comp-sci area with "not invented here" at every level and the only good people there were good people before coming in - but I will contain myself on this front just trust me that I have seen a few areas and can actually compare (to know a good vs bad one) and it's not some personal thing, for example the maths and physics departments are very good.

    The "critical being called normal" thing is way too high, it says 1% or 5 out of nearly 400 - most now (even in rural areas when I go an visit family) are digital machines so the technician actually sees the xray and clicks send or something - surely a "what do they think?" could improve it here, even if they're only trained in "normal looking" - it'll be something that can agree or not with a second source.

    There's actually some good mathematics behind efficiently analysing large sets of samples which cannot be pooled (eg say you're testing pee for something rare that when mixed with "chemicals" and shit glows bright pink, you can just pool (some of - not all the sample!) 10, 25, 50 ect piss samples together and test the lot - if the thing is sufficiently rare most wont turn pink at all, and thus you can tick off 10, 25, 50 ect as being "not my problem" - *you can't do this here* obviously)

    If anyone cares ... yeah... no...

    1. EBG

      Re: Heh it's Warwick

      It's Universities, amigo. Or more correctly the funding bodies. The behaviours that they dictate at grant application time, and demands to spin Impact in ways that would make Alistar Campbell blush. This whole Reg post is archived for my " Confessions of a KE professional" book after I retire.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2019