When IBM’s Watson supercomputer took on two human champions of the television quiz show Jeopardy and won, it was hailed as a breakthrough in machine intelligence. Now in an effort to expand the practical applications for the "world’s smartest computer," IBM Research and has taken the wraps off two new projects aimed at the medical community.
Developed in collaboration with the Cleveland Clinic Lerner College of Medicine of Case Western Reserve University in Ohio, the projects dubbed WatsonPaths and Watson EMR Assistant aim to find ways in which Watson can collaborate with doctors to improve training and clinical diagnosis.
NEW ATLAS NEEDS YOUR SUPPORT
Upgrade to a Plus subscription today, and read the site without ads.
It's just US$19 a year.UPGRADE NOW
The last thirty years have seen incredible advances in the field of medicine, but medical miracles aren’t of much use if there aren't doctors to perform them. According to the Association of American Medical Colleges, the demand for doctors in the United States in 2025 will outstrip supply by 90,000 physicians. If nothing else, this indicates that educating new doctors will need to be more efficient, and IBM Research believes that Watson may be part of the answer to training the doctors needed in the coming decades.
Costing a minimum of US$1 million, Watson isn't so much a single computer as a cluster of 90 IBM Power 750 servers containing 2880 POWER7 processor cores and 16 terabytes of RAM. It’s most famous for its turn on Jeopardy, where it outperformed two human opponents. After that, it went to university, did a stint at a cancer lab, and even got a job in customer service.
But machines like Watson have a considerable gap to cross before they can make the transition from quiz show contestant to everyday tool. Winning Jeopardy was a real feat as Watson used its ability to understand natural language and analyze unstructured data to answer questions. However, like chess, a quiz is a game that lives in a self-contained world of defined rules and Watson was in competition with the other players for a predefined end. This made Watson’s role adversarial, which is a fairly simple position, and winning was largely a matter of understanding the rules and developing strategies to achieve victory.
Medicine, on the other hand, is a collaborative task. Watson isn't meant to "beat" the doctor, but to work with him to find an answer to a problem. More important, it can’t be an answer delivered as if from a mysterious oracle. Instead, the machine has to provide its reasoning in a transparent, easy to follow form.
IBM Research entered into a year-long collaborative project with the Cleveland Clinic to find ways to merge the information processing and deductive abilities of Watson with the strengths and needs of clinical diagnosis. Doctors are very well educated, but computers aren't their area of expertise. This, paradoxically, presents a problem because these are people who are both well educated, yet very ignorant in a particular area, which means that the technology has to be much more user friendly than it would for the average consumer, if it isn't to generate a degree of professional resentment.
"On Jeopardy it was not necessarily critical to know how Watson arrived at its answer,” says Eric Brown, IBM Research Director of Watson Technologies. "But doctors or domain experts in any field will want to understand what information sources Watson consulted, what logic it applied and what inferences it made in arriving at a recommendation. Through our research collaboration with Cleveland Clinic, we've been able to significantly advance technologies that Watson can leverage to handle more and more complex problems in real time and partner with medical experts in a much more intuitive fashion. These are breakthrough technologies intended to assist future versions of Watson products."
WatsonPathsWatsonPaths is an expert system tailored to match clinical diagnosis in a way that is more natural for doctors. It uses a series of questions to gather information and to filter out possible alternatives. This is combined with data gathered from medical literature and what Watson has learned from doctors as feedback to draw inferences, and the result then displays the machine’s reasoning in a transparent manner using a graphic chart.
The reasoning path is designed to be very similar to that people use in real life. The idea is to ensure that it's relatively easy to understand where the information came from and how it’s applied, which is very important not only in diagnosis, but in physician training. One important factor of WatsonPaths is that it isn't a static system waiting for updates, but is a learning program that uses feedback from doctors to produce and evaluate new chains of reasoning that IBM claims will be “smarter.”
When WatsonPaths is ready to be let loose into the real world, it will be used by the faculty and students at the Cleveland Clinic in the problem-based learning curriculum and in clinical lab simulations. According to IBM, WatsonPaths will help in not only improve students’ diagnostic skill, critical thinking and problem solving abilities, but also in having a better understanding of the current medical literature.
"WatsonPaths is designed to augment the problem-based learning methods that Cleveland Clinic medical students employ in the classroom," says J Eric Jelovsek, MD, MMEd, Director of the Cleveland Clinic Multidisciplinary Simulation Center. "The vision is for WatsonPaths to act as a useful guide for students to arrive at the most likely and least likely answers to real clinical problems, but in a classroom setting. Of course, it is also easy to visualize how this type of technology could eventually be a tool for physicians to use in real-time clinical scenarios – a powerful guiding reference to consult when diagnosing and identifying the best treatment options."
Watson EMR AssistantOne of the most powerful tools in a doctor’s arsenal is the patient’s medical record. One diagnosis is of only limited value because it only provides a doctor with a snapshot of a person’s health. However, a complete record of illnesses, diagnoses, and treatments can provide insights into the patient's health that can tell the difference between an irrelevant incident and a vital clue. The problem is, the more information that's gathered, the better the diagnosis, but also the more data that needs to be sifted through. A lifetime of such data can amount to 100 MB.
That is where Watson EMR Assistant comes in. Its purpose is to process electronic medical records, allowing for discrepancies, and using its knowledge of natural language and semantics to understand the context of unstructured data and drawing relationships that might be of use to the doctor. If it succeeds, it should be able to provide a problem list of clinical concerns, draw attention to key lab results and medication that correlate to the list, and generate a patient timeline.
IBM and Cleveland Clinic recently discussed the role of Watson in medicine during the Cleveland Clinic Medical Innovation Summit, which ran from October 14 to 16.
The video below describes the features that WatsonPaths and Watson EMR Assistant uses.