Follow Slashdot stories on Twitter

 



Forgot your password?
typodupeerror
AI Medicine Biotech Businesses IBM

IBM's Watson Used In Life-Saving Medical Diagnosis (businessinsider.co.id) 83

"Supercomputing has another use," writes Slashdot reader rmdingler, sharing a story that quotes David Kenny, the General Manager of IBM Watson: "There's a 60-year-old woman in Tokyo. She was at the University of Tokyo. She had been diagnosed with leukemia six years ago. She was living, but not healthy. So the University of Tokyo ran her genomic sequence through Watson and it was able to ascertain that they were off by one thing. Actually, she had two strains of leukemia. They did treat her and she is healthy."

"That's one example. Statistically, we're seeing that about one third of the time, Watson is proposing an additional diagnosis."

This discussion has been archived. No new comments can be posted.

IBM's Watson Used In Life-Saving Medical Diagnosis

Comments Filter:
  • by Anonymous Coward

    Reads like an IBM ad.

  • Enough to see all these thinly-veiled ads, but it'd be nice to see Watson be proved wrong.

    • by skids ( 119237 ) on Sunday December 11, 2016 @10:34PM (#53466633) Homepage

      1) Why? Why would it be nice?
      2) Since it is an expert system, it probably already has an error rate. That's how expert systems work. They generate probable leads which feed into a diagnostic process as one contributing input. All IBM needs to prove here is that the leads a probabilistically worth the cost to generate them and the cost to follow up on them.

      • Why? Why would it be nice?

        To see the fallibility of Watson. Just as it was nice for Patterson to observe the original Watson seeing his own desk on fire, it is similarly nice to see Watson fail.

        Since it is an expert system, it probably already has an error rate.

        Yet we have not seen it in any final results. On the other hand, we've seen Watson's PR-friendly "successes".

    • by ShooterNeo ( 555040 ) on Sunday December 11, 2016 @11:27PM (#53466781)

      I'm sure it gets it wrong lots of times, but so do humans. It's only a problem (with Watson) if humans wrongfully assume it to be infallible or if Watson gets it wrong so often it's useless. This does not seem to be the case.

      • by nbauman ( 624611 )

        It's only a problem (with Watson) if humans wrongfully assume it to be infallible or if Watson gets it wrong so often it's useless. This does not seem to be the case.

        That's the question: Does Watson get it wrong so often it's useless? And do people wrongly assume it's infalliable?

        The only way to tell is to do a large randomized, controlled trial in which half the doctors are using Watson and half aren't, and see whether the doctors with Watson have better outcomes than the doctors without Watson.

        That's the way they study new imaging techniques (which is how they know a lot of imaging is useless).

        When they publish a randomized, controlled trial in a major medical journal

    • by DeVilla ( 4563 )

      Define "wrong". Watson might make 10 recommendations. For each it provides the information, references and "reasoning" it used to come to the conclusion. In the end the doctor always makes the choice. When a doctor doesn't take the top recommendation and goes with say recommendation 2, you ask the doctor what Watson had wrong and add to Watson's reference information.

      The important point though is the doctor is still the one who decides the diagnosis. Watson is just a tool to help find more pieces of i

  • DNA Sequencing (Score:3, Informative)

    by Anonymous Coward on Sunday December 11, 2016 @09:49PM (#53466443)

    Leaving aside IBM plugging Watson, which is just them trying to market a brand into healthcare, the actual Leukemia detection was DNA sequencing with abnormalities correlated to research papers.

    That in turn suggested she was pre-disposed to Myelodysplastic cancers, and needed blood transfusions, and other support drugs.

    It looks like IBM wants to perform the middle-man deception on an epic scale here. Where the 'expertise' is being presented as IBMs when in fact its just searching others research papers with a not-very good language parser to grab references to the gene from the documents.

    • I fully agree. The entire gene analyzing startup industry does this though, like 23andme or others. They are basically grep inc.

    • Re:DNA Sequencing (Score:5, Insightful)

      by Comrade Ogilvy ( 1719488 ) on Sunday December 11, 2016 @10:22PM (#53466581)

      What Watson did is something any intern with time on his or her hands could have achieved by looking at the data, searching in a medical database, and then a long bit of drudgery finding the wheat among the chaff. And if it were an intern, we would applaud that person for surprising tenacity and "vision" in helping a patient when seasoned doctors failed.

      Why do you need to denigrate that? Why do you need to concoct a phony baloney different standard when a computer succeeds where humans failed?

      • Re: DNA Sequencing (Score:3, Insightful)

        by Anonymous Coward

        I think some people just reduce all problems to a basic summary, then once they can feel superior by letting everyone know that if it's just such a basic thing they could do it in a few lines of perl that they're clearly winning at life and these other engineers are just faking it and deserve zero credit.

        It's a way of masking their own insecurities.

      • by Kjella ( 173770 )

        What Watson did is something any intern with time on his or her hands could have achieved by looking at the data, searching in a medical database, and then a long bit of drudgery finding the wheat among the chaff.

        And perhaps the important part is that the wheat:chaff ratio is only getting worse as we get more and more knowledge about increasingly obscure conditions. Practically we can only ask doctors to learn this much before they have to start practicing medicine. Practically we don't have time for interns to chase needles in a haystack very often.

        Sure, in theory you could boil it down to a theory of indicators and counter-indicators combined with tables on base incidence and prevalence, modifiers for age, sex and

      • by nbauman ( 624611 )

        We're denigrating the way David Kenny was making vague claims rather than giving us the critical, specific facts that we need to tell whether Watson is a new toy or actually useful. We're denigrating Kenny for talking like a salesman, not like a scientist.

        Statistically, we're seeing that about one third of the time, Watson is proposing an additional diagnosis.

        That's great. Statistically, how often does that additional diagnosis turn out to be correct, how often does that additional diagnosis change the way the patient is treated, and how often is the outcome any better?

        A British radiologist working in the US de

      • by nbauman ( 624611 )

        That's right.

        I expect that an experienced oncologist could have done the same with the information in his brain, without the drudgery.

        All he has to say is, "Let's order that leukemia DNA test panel."

      • " and then a long bit of drudgery finding the wheat among the chaff."

        Drudgery is exactly the kind of thing that computers are good at and humans aren't.

        A human researcher might have made the discovery, but it would probably take so long the patient would already be dead.

        Yes, the original article is an IBM marketing piece but this kind of advanced greppery is likely to be the norm in a few years.

    • by nbauman ( 624611 )

      You got it. Out of mod points, sorry.

  • .. did not know it would pop up so prominently on /.
  • by Comrade Ogilvy ( 1719488 ) on Sunday December 11, 2016 @10:17PM (#53466557)

    My personal opinion is most doctors are probably pretty bad at diagnosing non-obvious issues. We do not actually need Watson to replace the doctors. We need Watson as another opinion who looks at the data in another way, and can usefully point to the long tail of uncommon to rare things that have a statically reasonable likelihood of being relevant. Many of these uncommon things, why would expect a doctor to actually be competent at diagnosing them? When would they have built that kind expertise?

    Taking TFA at face value, the doctors were ignoring data right under their noses. Watson found it by simply looking. It is not a matter of Watson have some magical genius. It is a matter of Watson being simply and thoroughly competent at many, many easy things that most doctors can never be expected to learn.

    • My personal opinion is most doctors are probably pretty bad at diagnosing non-obvious issues. We do not actually need Watson to replace the doctors. We need Watson as another opinion who looks at the data in another way, and can usefully point to the long tail of uncommon to rare things that have a statically reasonable likelihood of being relevant. Many of these uncommon things, why would expect a doctor to actually be competent at diagnosing them? When would they have built that kind expertise?

      Taking TFA at face value, the doctors were ignoring data right under their noses. Watson found it by simply looking. It is not a matter of Watson have some magical genius. It is a matter of Watson being simply and thoroughly competent at many, many easy things that most doctors can never be expected to learn.

      In other words, we DO actually need Watson to replace the doctors, which you have quite clearly demonstrated to be incompetent of providing accurate diagnoses that include ALL variables due to the overall complexity today.

      By the way, this concept should be familiar, as years ago computers started replacing human work. It only makes sense that evolution continues where complexities are well beyond a human capability.

    • I think the more realistic reason is that a doctor simply can't know every single rare condition off the top of his or her head, and many rare conditions share symptoms with common conditions, making it easy to overlook. You also might be surprised how often doctors who pop out of the room are sneaking a peak into a text book to confirm treatment.

      That's why diagnosis might change going to a specialist (who is going to be more familiar with rare conditions).

      Additionally, I see no harm in adding in an additi

  • How much does it cost?

    I'm asking because it cost me $4,000 out-of-pocket to merely identify if my uncommon, congenital disorder is the really bad kind or just the somewhat inconvenient kind (it was the latter).

    Doing a differential analysis must cost a boatload of cash on top of this.

    • How much does it cost?

      Wait 18 months. If it's not cheap enough "out-of-pocket", then wait another iteration until the tech is ubiquitous.

      How much does it cost?

      And for fucks sake, get your country single payer, universal health coverage already.

      • I, too, want Donald Trump to pick who is in charge of the health system. What could possibly go wrong? And just think about the endless arguments about whether the single payer should cover things like birth control or abortions. Really though, isn't it just time for the government to openly declare us subjects, and quit pussy footing around pretending it doesn't own us, mind, body, and soul?

  • by Joe_Dragon ( 2206452 ) on Sunday December 11, 2016 @10:53PM (#53466681)

    "You've got leprosy, goodbye."

  • Leukemia is not a virus, or an animal or plant species...
    Perhaps they mean "types" of leukemia, but it is hard to imagine that her docs missed that. Oh and so they just decided to "treat both strains" of leukemia and now she's great! something doesn't sound right here...

    • Cell lines can't have genetically determined strains? Not that I'm a biologist but it makes sense to me. "Types" definitely do not seem to be the thing in question here. My impression is that the genetics of those cells has gone bad in two different ways separately from each other or something like that.
    • Shitty reporting doesn't mean facts aren't facts. Be suspicious after reading the peer reviewed paper, or a statement directly from the team. Critique of a Dashslot summary is no art form...

  • N=1, no details (Score:5, Insightful)

    by nbauman ( 624611 ) on Monday December 12, 2016 @02:02AM (#53467255) Homepage Journal

    There's a 60-year-old woman in Tokyo. She was at the University of Tokyo. She had been diagnosed with leukemia six years ago. She was living, but not healthy. So the University of Tokyo ran her genomic sequence through Watson and it was able to ascertain that they were off by one thing. Actually, she had two strains of leukemia. The did treat her and she is healthy.

    This is one anecdote by David Kenny, General Manager of IBM Watson, who has no medical expertise. He's a computer salesman and this is his pitch. Have you ever heard of a computer salesman making a promise that turned out to be an exaggeration?

    This is the kind of story I used to see about miracle cures from vitamins or fad diets or Burzynski's Clinic: a vague description ending in a miraculous cure that is impossible to verify.

    Kenny doesn't even describe the case. She had "leukemia." What kind of leukemia? There are at least 4 major types of leukemia, and many subtypes.

    Some of them have effective treatments, some of them don't.

    Some of them have a median survival of 6 months, some of them have a median survival of 20 years. Since she's been alive for 6 years she doesn't have the most deadly type.

    It looks as if they found a mutation which suggested that one drug would be more effective than another, which is routine these days. You don't need a supercomputer to do that. You do need a randomized, controlled trial to see if using Watson actually leads to better survival than doctors can get without Watson.

    She's healthy? What does "healthy" mean? What does "not healthy" mean? Those aren't medical terms.

    If they're serious about this, publish in a medical journal.

    • Ever look up something with an internet search? How to fix a leaky sink or overcome a coding obstacle... This is the supercharged version of computer assistance.

      I've been in the doctor's office and had a nurse practitioner google my symptoms. I am a big fan of diagnosticians who are unafraid to ask someone if they don't know.

      • by nbauman ( 624611 )

        Have you ever looked up a disturbing symptom on the Internet? Have you done it not just once, but several times?

        If you do a thorough search, you'll find that most symptoms have a differential diagnosis that includes cancer. In other words, the problem with Google searches is massive false positives.

        There's a medical student's joke: If your doctor asks you what's wrong with this patient, say, "It's either cancer, an autoimmune disease, or an infection." The joke is that most symptoms could be either cancer,

        • Except that you are wrong about your own diagnosis. I have a friend who's a former Olympic rower, and competes in Senior class to this day. About 15 yrs ago he got up one morning and felt extremelydizzy and weak. Med exam revealed a heart valve failure.
          So that's nice you ran up X flights of stairs. It doesn't preclude a heart problem.

          • by nbauman ( 624611 )

            If he had a heart valve failure, then he wouldn't be able to walk up 5 flights of stairs.

            My broader point is that the internet is a great additional resource for medical education (it won't replace textbooks and journals, much less lectures and labs).

            But at the end, you need a doctor (sometimes more than one) to decide what the problem is.

            There is a movement now to save money by eliminating doctors and delegating much of their job to nurse practitioners and even computers. They're trying to turn patients in

  • What happens when Watson comes to the conclusion that single payer health care would save more lives than the current US health system?
  • by QuietLagoon ( 813062 ) on Monday December 12, 2016 @10:09AM (#53468717)
    IBM is crowing about saving peoples' lives. Microsoft is crowing about what is probably a couple of people switching from Mac to Surface.
  • "...ran her genomic sequence through Watson..." could we maybe get a little detail on that? its sort of a complicated thing to do with a ton of variables

Who goeth a-borrowing goeth a-sorrowing. -- Thomas Tusser

Working...