Artificial Intelligence

Watson helps make diagnoses

Diagnosing rare diseases is the specialty of Jürgen Schäfer at Marburg University Hospital. Now the German “Dr. House” is getting help from IBM's cognitive assistant Watson.

09 Nov. 2016 Jens Stark
IBM-Watson-01
Jürgen Schäfer diagnoses rare diseases with the help of IBM Watson. (Photo: Pavel Chagochkin /Shutterstock.com)

Dr. House and Dr. Watson

Jürgen Schäfer could be viewed as a German "Dr. House": Like the U.S. television character, he and his team are specialized in rare diseases. He has been treating patients at the Marburg University Hospital's ZUSE Center for Unknown and Rare Diseases since 2013, some of whom have gone from one doctor to the next for years or even decades without receiving a satisfactory diagnosis of what was wrong with them. “Many of our patients have a long medical history and bring us heavy piles of documentation from earlier examinations,” reports Schäfer. Finding the missing link that leads to the correct diagnosis is something of a needle in a haystack challenge.

IBM's "Dr." Watson

But now the ZUSE doctors have in IBM's Watson cognitive system a digital assistant that can help them explore the causes of these rare diseases more quickly, and reduce the waiting list of now 6,000 people seeking their insight, as was announced at a media event for IBM Research – Zurich in Rüschlikon, Switzerland.

The Marburg team tested Watson's abilities during a twelve-month pilot project, by feeding the system material from already solved cases and assessing the diagnoses reached by the digital “doctor.” Apparently Watson's performance convinced the ZUSE doctors, who now want to use the system for open cases.

How Watson does it

Here is how Watson works: First the system conducts a text analysis of the patient documentation, using natural speech analysis to not only isolates symptoms, but also “understand” what medications have been tried. And it does not stop there: The system also extracts from the medical history the diagnoses that were ruled out, and the medications the patient does not tolerate.

Along with this analysis, Watson evaluates its available knowledge base, which is made up of different sources such as medical anthologies and databases, as well as Wikipedia entries. “Watson is constantly being fed new medical publications,” says Matthias Reumann of IBM Research – Zurich, pointing out that two medical articles a minute are being published these days.

Along with this knowledge base, the ZUSE physicians have developed an online survey of 1,200 questions that provides structured information.

Will doctors become obsolete?

Once all this has been input, Watson's “cognitive core” gets to work, matching symptoms to diseases, for example, or vice versa. At the end the system presents the doctor with a range of possible diagnoses, which are weighted into a ranking. Reumann assures com! professional that Watson's diagnoses are very transparent and can be traced and reproduced by the medical professionals.

Watson is a useful tool, but no replacement for a doctor

Schäfer at ZUSE does not yet fear for his livelihood. He sees Watson as a capable assistant who cannot replace his own abilities, nor reduce his patients' confidence in him. He compares his situation to that of a pilot, who naturally relies on the data from her onboard computer without passengers losing faith in her skill. On the contrary: “What would they say if the captain on a New York flight announced that he was planning to turn off the onboard computer to improve their confidence?” Schäfer asks rhetorically.

Artificial Intelligence CEBIT RSS Feed