June 01, 2011
Winning the $1 million "Jeopardy!" challenge earlier this year was just the tip of the iceberg for Watson, the IBM supercomputer that experts believe has the potential to revolutionize the healthcare industry.
Well before Watson's gameshow victory, experts at the University of Maryland (UM) School of Medicine in Baltimore and Columbia University Medical Center began working with IBM to apply Watson's analytics capabilities to healthcare. Specifically, Watson is being developed as an assistant capable of reading electronic health records (EHR) and providing instant feedback to physicians in ways not always available from doctors and nurses.
"This breakthrough in computer science will allow us to explore this technique for medical diagnosis," said Eliot Siegel, M.D., a professor and vice-chair of imaging informatics at the University of Maryland School of Medicine (MSM), chief of imaging services for the Maryland Veterans Affairs (VA) Healthcare System at the Baltimore VA Medical Center and co-chair of RSNA's Medical Imaging Resource Center (MIRC) committee.
"The potential for a renaissance in electronic health records really lies in the evolution of computer systems," said Dr. Siegel, director of UM's Maryland Imaging Research Technologies Laboratory, who was instrumental in the partnership between MSM and IBM. "I'm really surprised it has taken this long for that renaissance to start."
Physicians at Columbia University are helping identify critical medical issues to which Watson may be able to contribute, according to IBM.
Analytics Capability Critical to Healthcare
Powered by 90 servers and 360 computer chips, Watson was built in four years by IBM researchers seeking to develop a machine that could quickly answer complex questions. Through IBM's Deep Question Answering, Natural Language Processing and Machine Learning statistical techniques, Watson works to understand questions and develop answers—a capability critical to the technology's potential value to healthcare.
IBM is also working with speech-recognition software developer Nuance Communications to give Watson the analytics capabilities necessary for physician-patient consultations.
Earlier attempts at artificial intelligence required every possible question and answer to be hard-coded into the system, a time-consuming process with little value in healthcare, said Martin Kohn, M.D., Chief Medical Scientist, Care Delivery Systems, IBM Research.
"Watson uses a probabilistic, evidence-based approach," Dr. Kohn said. "It generates and scores many hypotheses using an extensible collection of natural language processing, machine learning and reasoning algorithms. Many previous such efforts relied on programmed decision rules. Watson is a self-learning system that does not rely on such rules. It gathers and weighs evidence to refine its hypotheses."
Decision Support Boosted to New Level
Radiology stands to benefit tremendously from Watson's capabilities, experts say.
"The technology has the potential to provide decision support on a scale not dreamt of prior to this," said Nancy Knight, Ph.D., the director of Academic and Research Development and a founder of the Maryland Imaging Research Technologies Laboratory at UM.
"Watson can supply the radiologist at the point of care with complete patient information from the electronic health record, including imaging history, allowing the radiologist to mine an often exhaustive number of records to identify the most important points," Dr. Knight said. "It also provides the latest and most extensive scientific knowledge and clinical experience that can be used to inform decisions about diagnosis, additional tests, management and likely prognoses."
Watson is currently in the testing phase in that learning process, said Dr. Siegel, who pointed out the similarity to real-life students progressing from medical school to residencies.
The first step—acquiring book knowledge—is already under way. Watson's database already includes information from medical journals and textbooks such as the Merck Manual of Diagnosis and Therapy, Harrison's Principles of Internal Medicine, the American College of Physicians Medicine and Stein's Internal Medicine.
Next, experts will work to develop Watson's understanding of the physiology of the human body, followed by the third step: gathering experience.
"Watson not only needs the general knowledge that made him so successful on 'Jeopardy,' but also information from the databases specific to medicine," Dr. Siegel said.
Watson is an Assistant, not a Physician
In time, Dr. Siegel would like to see Watson function as a physician's assistant. He envisions Watson being used for chart review, providing assistance on drug interactions or inconsistencies in prescriptions. Regardless of the technology's potential, Dr. Siegel stresses that Watson is designed to act as an assistant to a physician, rather than a replacement.
"I don't see this technology supplanting physicians or radiologists," Dr. Siegel said. "It's a tool that will gather, summarize and analyze information—very similar to the role now performed by our best residents and fellows."
As the technology progresses and expands, it will be especially important to rural hospitals or areas where there may be fewer experts, but healthcare organizations across the board stand to benefit, he said.
"Watson's eventual expansion into healthcare has profound implications for radiology and will certainly improve the safety, effectiveness and potentially the cost of healthcare delivery overall."
Siegel Discusses Watson's Role at UM
To view a video podcast of Eliot Siegel, M.D., discussing the development of Watson as an assistant capable of reading electronic health records and providing instant feedback to physicians at the University of Maryland School of Medicine, click here.