When your dog hears a human word, he processes it his own way. This could be helpful for training. More >
Can a Computer Diagnose Disease? Researchers Say We're Getting Close
The supercomputer affectionately called Watson, after IBM’s first president, recently beat out two (human) champions on the TV game show Jeopardy! Many people, including researchers, the media, and the public, hailed Watson’s victory as a major step in robotics research: his incredible ability to understand – and respond to – all the subtleties of human language had not been seen before.
But can the computer be given a medical "education", and be programmed to respond accurately to complex medical dilemmas that might stump even the best doctors? Researchers at Columbia University’s College of Physicians and Surgeons have teamed up with IBM to give Watson its medical training — and they say the answer to this question is "yes".
Herbert Chase, MD is a researcher and professor in clinical medicine at Columbia’s Department of Biomedical Informatics. He also acts an advisor to IBM, supervising and rigorously testing Watson’s medical programming. We spoke to Dr. Chase about Watson’s medical "training" and the role the supercomputer will play in the future of medicine.
Chase tells us that unlike the Jeopardy! judges, "physicians are not looking for a single best answer." When doctors begin to diagnose disease, they come up with a list of the several most likely candidates, which is known as "differential diagnosis". Then they whittle the possibilities down until they arrive at the right one — this process, as you might guess, can be difficult and time-consuming, even for the most experienced specialists.
A medical Watson could be a big help in the process "by providing physicians a more complete ‘differential diagnosis’." Watson will be able to come up with possibilities that the doctor may not have thought of him- or herself, which is especially important in the diagnosis of a rare disease. As Chase reminds us, "patients do, with remarkable regularity, present with very rare conditions."
The mind-boggling amount of scientific information that exists these days is the main reason why a supercomputer like Watson would be a great benefit to doctors. "For decades," Chase says, "it has been appreciated that physicians can not only not master the information required to provide the best care possible, they can’t even access it when given sufficient time. The quantity of information that informs care is so vast, that only sophisticated computer programs can access, filter, and identify useful and important information."
After Watson’s medical programming and reprogramming, the supercomputer will be able to access the enormous amount of available scientific information that a human doctor could never realistically tackle. Doing so will help doctors arrive at more accurate and timely diagnoses for patients with a wide range of conditions.
Watson may also help cut down on medical error. Chase explains that about 25% of medical error comes from "delayed, missed, or incorrect diagnostic error."A doctor can use Watson to search the medical information in an individual patient’s Electronic Health Records (recently mandated by the Federal government) and then suggest diagnoses even though the physician did not enter this information into the machine. He or she can also directly enter a patient’s information into Watson and see what diagnoses are suggested. Or, Watson could extract information from the patient’s Electronic Health Records, without the physician’s input, and identify potential diagnoses."
Watson is already diagnosing medical conditions like a pro. "Even now, after little real effort, the machine is answering questions with remarkable accuracy," says Chase. His job is to determine how Watson is doing. He goes over the star computer's answers and suggests ways to improve its accuracy. He is also testing how "usable" a medical Watson will be. Specifically, he and his team are exploring just how the supercomputer will result "in a change in work-flow", as well as looking at the "optimal way that a user would actually use the machine during a clinical encounter".
So where will a medical Watson reside in a healthcare facility — will a doctor schlep a tiny Watson in to the exam room, or will it stay behind the scenes, acting more like an assistant or ghost-doctor? It is likely the latter scenario that will be the case, with Watson acting like a behind-the-scenes medical consultant with which doctors can confer after seeing a patient.
Chase underlines that a medical Watson would exist to provide information, not make final diagnoses: "Watson’s role is to provide the most up-to-date and complete information which the physician can use to inform his/her practice. Watson will not be making decisions. In fact, we teach our students that it is the patient who makes the decisions regarding his/her health." Chase adds "the physician’s role [is] to fully inform and explain to the patient all the relevant information (risks, benefits) so that the patient can come to the best decision. It is Watson’s role to provide information to the physician to keep the physician up-to-date."
Time will tell how Watson’s medical "training" will progress, but from the early results, it looks like it will offer the medical world some major advantages. Though the idea of a machine understanding personal symptoms and suggesting possible diagnoses may seem strange, it’s likely that a medical Watson will revolutionize medicine and offer significant benefits to patients dealing with a range of conditions.
April 1, 2011