Supercomputers can be handy medical assistants, IT experts and doctors agree. But not every hospital needs one as smart as IBM’s Watson.
IBM Corp. has recently offered the services of one of its most famous high performance computers to Cedars-Sinai Medical Center in Los Angeles. Watson is lauded for its ability to provide speedy diagnostic advice, reducing human error and hopefully, cutting down on the number of medical malpractice suits down the road. At Lotusphere 2012 in January, Manoj Saxena, general manager of IBM’s Watson group, regaled the crowd with some of its exploits.
But if you’re a hospital and Watson is outside of your price range, don’t fret, say experts. These days, a “supercomputer” can cost as little as $10,000. The term itself is a relative one, referring merely to the highest-performing systems at any given time. And in fact, many hospitals are already using systems that can do some pretty heavy lifting. Sure, they might not be as bright as Watson, or have quite the celebrity status, but they’re good at what they do.
A better biopsy
Speaking of Watson, Dr. Andrew Evans, director of Toronto’s University Health Network
telepathology program, says the infrastructure at his hospital group includes “absolutely nothing to that degree of sophistication.”
To be sure, telepathology at UHN does demand some serious computing power. Take, for example, the digitization of microscope biopsy slides, transmitted from institution to institution over the Internet (sometimes across the country), which produces massive amounts of data.
“The digital slide files are huge,” says Evans. “One typical microscope slide from a major surgical specimen like, say, a man has his prostate out — one section from a prostatectomy case could be 1.2-1.5 GB. That’s just one slide. And then that whole specimen could have 60 of those slides.”
Doctors can view and zoom into these images on large computer screens. “It replicates flipping the nosepiece on a microscope and going to higher magnification,” he says.
Computer-aided diagnosis, which Evans says is not yet in place at UHN, would require even more serious backend power, machines that would need something close to Watson-type smarts. “For example, if you have a biopsy slide from a man’s prostate, could you train a computer to recognize prostate cancer?”
Supercomputers are suited to this kind of work because of the complex contextual information inherent in the diagnostic process, says Steve Conway, research vice-president of high performance computing at IDC
. “It’s not something where the situation is the same for every patient and you just go, ‘Oh, somebody just walked into a wall. What should we do to deal with the bruises?’” he says.
“A lot of these situations for diagnosis may be in conjunction with surgery and so forth, where every human being’s history and every human being’s anatomy is different. You really need to take that into consideration. It gets to be pretty complicated.”
Finding a disease, finding a terrorist
But Israel Gat, director of Cutter Consortium’s
agile practice, questions how complicated things need to get in medicine. He draws a line between “macro” and “micro” uses of high-performance computing in hospitals.
“Yes, high performance computing, and big data and modern analytics, etc., should be phenomenally good for macro trends in health care. For example, if you want to know what the prospects are that swine flu would evolve in Malaysia, or whatever, then this kind of analysis, what we often call digital reasoning, is absolutely fantastic.
“For example, the intelligence community is using it for security and warfare purposes. And at least in my opinion, the same would apply, as is, to health care. The only difference is that instead of looking for an al-Qaeda contact, you look for swine flu, or whatever disease you are looking at, amongst billions of patients.”
On the other hand, he says, in a single hospital or clinic, as opposed to 10,000 institutions around the world, he doesn’t see a particularly compelling case for investing in “super power.”
“My understanding of the medical profession, and it’s probably a layman’s, is that basically they have got a pretty good idea in terms of what Doctor X or Dentist Y is doing [and] about what needs to be done.”
A professional who has to decide between two strategies could, of course, rely on a computer for help, says Gat. He just isn’t convinced that a doctor needs an especially brilliant sidekick.
“Let’s say we take, for example, a dental procedure. Maybe they want to do an implant, maybe they want to do a crown lengthening. But the pros and cons between one and the other are fairly well known. At the micro level, namely if I am a dentist and I need some quick advice, probably it [requires] no more than encoding knowledge that they already have at the state of the art.”
Dr. Watson, department of pathology?
On having a supercomputer as a colleague, “it would be fantastic,” says Evans. “Think of the possibilities of a Watson supercomputer, teaching it how to interpret digital data.”
While he says he welcomes computer help for diagnoses, other pathologists are less comfortable with the idea. But it’s not so much the Watsons that frighten them.
“It is fair to say that pathologists are threatened. Some are threatened. And I don’t mean to trivialize it in any way, but they’re threatened by the digital pathology technology on a number of levels,” Evans said.
For one, as a generally conservative bunch, they worry that if new technology is pushed on them too quickly they could make costly mistakes. There’s also a fear that new advances could lead to the diagnostic process being outsourced to some remote locale. “In theory, you could get a group of pathologists in India who would charge a fraction of the price to review all these cases. And there goes pathology in Canada,” says Evans.
But Evans says we shouldn’t expect computers, even highly intelligent ones, to replace the judgment of a trained pathologist. If computer-aided diagnosis does see use on a broad scale, he says, its value would lie in providing a second opinion of sorts. A computer’s interpretation of what it sees in a microscope slide could reduce the variability in what several human pathologists see in it, or perhaps, how one pathologist perceives the same slide at different times.
“It’s very much a subjective specialty. And the whole aim of the computer-aided diagnostic side, which as I say is not online yet, is to try to tighten up some of that stuff,” he says.
“Ultimately, when it does come online it wouldn’t be something that would replace pathologists. It would just enhance the way pathologists do their work.”