Dragon Go!, the
consumer version for
Andriod phones, was
shown at CES 2012

Nuance talks bringing Watson, Siri-like tech to medicine

January 26, 2012
by Brendon Nafziger, DOTmed News Associate Editor
A doctor is about to prescribe an antibiotic, but is a bit worried it might interact with an antidepressant her patient is currently taking, Wellbutrin. So, what does she do? It's possible, in the near future, she might simply pull out her phone and ask it.

At the recently wrapped up 2012 Consumer Electronics Show, Nuance brought Android phone owners a program similar to Siri, the popular voice-command search feature released by Apple in its most recent iPhone rollout. The free app, Dragon Go!, is strictly for consumer use. But Nuance, which licenses some of its technology for Siri (Nuance and Apple are pretty closed-lipped about specifics), also has a medical version in the works.

A prototype of the as-yet-unnamed app will make its public debut at HIMSS 2012 next month, but it will be a quiet one.

"We'll carry a few of them around in our pockets," Joe Petro, senior vice president of research and development with Nuance Healthcare, told DOTmed News.

Nuance made its name in speech recognition software. And the product, which Nuance plans to release commercially later this year, takes advantage of what Petro calls the "next evolutionary step" in the technology.

"We think we have the speech dictation down. We're chewing at the upper edge of accuracy, getting from 96 to 99 percent," he said. "Most of the stuff you see from an innovation perspective in 2012 will be about exploiting that understanding. From the understanding perspective, it's almost untapped."

Language processing

In essence, the technology would work by breaking down spoken language, categorizing it and mapping the categories onto a series of medically oriented databases - for instance, for drug interaction (as in our antibiotic and antidepressant example) - and then returning a context-specific result.

As it acts more like a reference book than a computer-aided diagnosis tool, the technology likely wouldn't fall under the Food and Drug Administration's purview. Still, the company hopes doctors can find it helpful. The drug interaction use case is, in fact, rather important here, as Petro said Nuance sees a lot of potential for this technology in helping doctors navigate the perilous waters of pharmacology.

"One of the big challenges [for doctors] is keeping up with all the pharmaceutical stuff," he said. Treatment protocols and dosing information are frequently updated and quite complex, so having a ready reference to aid physicians in prescribing drugs to someone already taking others could be useful.

Of course, this feat, rapidly and intelligently finding answers to technical questions, bumps up pretty close to the turf of another much-ballyhooed technological figure from last year that Nuance also is working on: Watson.

The Dr. Watson dilemma

Watson, originally called DeepQA when developed by IBM, can rapidly browse through a vast store of information. Like Johnny Five, the speed-reading robot in the 1980s comedy "Short Circuit," Watson works fast, and can page through about 200 million pages of data in three seconds, according to a presentation Nuance gave at the Radiological Society of North America's annual meeting last fall.

Using all the data at its electronic fingertips, plus its hypothesis-testing programming that helps it puzzle out a user's "intent," Watson can then provide answers to specific questions. Famously, this was done answering trivia on the TV show "Jeopardy." But almost immediately, people began wondering if there were medical applications for it.

It turns out Nuance has been working with IBM (again, the company is cagey about specifics), doing research on using a Watson-like technology in medicine. Petro said when they originally approached IBM about using Watson, the first idea bandied about was essentially "Dr. Watson": having the software answer specific questions related to specific diagnoses. But Petro said they decided it could be an "unnatural interaction" for physicians, and that it could "tread too closely on what is ultimately their clinical decision-making." What doctors want, Petro suggests, is a consultative process, where they work with the software to get deeper and deeper levels of information about medical topics, such as common treatments for psoriasis.

"We call it spiraling," he said. "You can imagine spiraling down and converging on a single body of information to provide final answers to support the conclusion the physician has made."

As a test, Nuance said it's working on getting the technology proficient at a type of "medical Jeopardy" known as Doctor's Dilemma. Developed by the American College of Physicians, it provides a set of standard questions in medicine, like what is the most common vector of rabies in the United States (hint: it's a mammal with wings).

"[Doctor's Dilemma] involves all the critical pieces of getting the system to work, in terms of accessing a different variety of repositories, so we are internally benchmarking [the technology's proficiency] on a day-to-day basis," Petro says.

Right database, radiology

Feeding this technology the correct information is key, which is why Nuance said IBM is focusing on oncology, where there's a well-defined body of information the technology can rely on. Nuance said it's interested in setting up a repeatable, reliable process to refresh content to keep the technology up-to-date.

Although drug formulary issues are important, Nuance also believes there's a use case for the technology in radiology. The company said it's sticking with its strengths -- a future product likely wouldn't take into account imaging data -- but it could be fed reports or asked questions to help physicians get the information they need while interpreting images.

"There are whole radiology-specific content resources," Petro said.