ACR Bulletin - July/August 2011 - (Page 13)
WILL SUPERCOMPUTERS EVENTUALLY REPLACE RADIOLOGISTS?
By ByArunKrishnaraj,M.D.,M.P.H. his past winter, Watson, the supercomputing brainchild of IBM, wowed the world as it easily defeated two of the greatest trivia masters on Jeopardy! The surprising victory prompted many observers to pen articles examining the impact of this technology on the relationship between computers and humans. Writers in industries from health care to hedge-fund management commented on implications of such an event.1 One news story specifically mentioned radiology as an arena where artificial intelligence may replace humans.2 So, are we destined for the dystopian futures portended by such sci-fi classics as The Matrix and The Terminator?
Breaking the ‘Natural Language’ Barrier
To compete on Jeopardy!, Watson was programmed with vast quantities of knowledge on a range of topics as well as the ability to sort through those topics when presented with a clue. It should come as no surprise, then, that Watson did so well on a game show that requires a substantial knowledge base and the ability to access it quickly. Isn’t that what computers were designed to do? Computers can parse but not understand natural human language. They require the use of computer programming language to comprehend your directives. Watson, however, uses a natural language processing engine that has the ability to overcome the humancomputer language barrier. It can comprehend natural language, including puns, word play, and metaphors common in Jeopardy! clues without additional programming.
What impact will this type of disruptive technology have on the field
of radiology? As radiologists, we review numerous images and base our findings on our experience and expertise, which result from reading articles and textbooks that form the foundation of our knowledge base. If we program a computer with all of our knowledge, then wouldn’t that computer know more than we do and be more effective at making diagnoses? With the ability to understand natural language, a primary care physician could ask the computer following an abdominal CT scan, “Does my patient have appendicitis?” and the computer could answer, “No, but a high likelihood exists that the patient has diverticulitis.” The advancement of machine intelligence makes such a conversation distinctly possible in the near future. The power of such analytic tools as Watson to improve patient care should not be discounted. As radiologists, we must seek ways to use this type of technology to augment our ability to provide more value-added care. With this in mind, a team of clinicians and software developers within the Massachusetts General Hospital Informatics Division have leveraged advanced electronic health record technology to improve the quality of care in its imaging department through the development of a programmable, ontology-driven search engine named the Queriable Patient Information Dossier (QPID).3 By allowing for fast, concept-specific information retrieval, intelligent search engines, such as QPID and Watson, could improve imaging interpretation by integrating and organizing relevant clinical history into our clinical workflow. Natural language processing tools also enable automatic data structure and alerts if data required to meet a quality reporting metric are absent. Moreover, this type of technology can improve
safety in imaging by systematically and quickly searching the medical record for evidence of contrast allergies, duplicate exam orders, cumulative radiation dose exposure, or MRI-incompatible implants.
Obviously, radiologists should employ technology that increases the efficiency and assurance of safety. But in delivering quality care, the personal relationship between a doctor and his patient is, and will always be, paramount. Discovering cancer, for example, requires diagnostic skill but also compassion when communicating that finding. A computer will never be able to demonstrate true human concern or rest a hand of comforting support on the shoulder of a patient that is hurting. Radiologists must continue to foster personal relationships with patients and referring providers. We must also be honest. Computers now augment our imaging capabilities, but one day, computers will surpass us with faster and more accurate diagnoses. To distinguish ourselves from machines, we need to engage our patients and referring providers in meaningful discussions. If not, we should get used to working for our computer overlords. //
ArunKrishnaraj,M.D., M.P.H. (firstname.lastname@example.org), is a radiologist at Massachusetts General Hospital in Boston.
ENDNOTES 1. Martin, Keith L. “Trendspotter: Temporary and Tech Solutions to the Physician Shortage,” Feb. 17, 2011. Available at: http://bit.ly/fxj461. Accessed June 7, 2011. 2. “The Dark Side of Watson,” Feb. 20, 2011. Available at: http://n.pr/hWAjY4. Accessed June 7, 2011. 3. Zalis M., Harris M. “Advanced Search of the Electronic Medical Record: Augmenting Safety and Efficiency in Radiology.” JACR 2010;7:625–33.
Advocacy • Economics • Education • Clinical Research • Quality & Safety | 13
If you would like to try to load the digital publication without using Flash Player detection, please click here.