How is the medical profession using AI systems?

Do computers diagnose symptoms better than human doctors?
Do computers diagnose symptoms better than human doctors?

The television medical drama "House" was a big hit during the 2000s, and not just because of the trademark sarcastic wit that the curmudgeonly lead character, played by Hugh Laurie, wielded with such wicked deftness. Another key part of the show's appeal was watching the brilliant Dr. House and his nearly-as-talented team of diagnosticians work together as a sort of hive mind to investigate perplexing medical mysteries. They pieced together bits of information and debated their significance, until finally arriving at a solution and devising some ingenious treatment to save the patient.

That all made for fascinating drama. Wouldn't it be great, though, if your doctor could consult a diagnostician who was even more brilliant, knowledgeable and resourceful than Dr. House (and a lot less prickly)? And what if any physician could seek that assistance from the computer on his or her desk, with just a few keystrokes?


The good news is that doctors now can tap into that sort of expert help, thanks to a new generation of medical software programs that use artificial intelligence (AI) -- that is, the capability of computers to emulate and improve upon human thinking. AI systems have the potential to revolutionize medicine. While a human physician -- especially a tired one at the end of a long shift -- might overlook or misinterpret the subtle symptoms of a rare disease that he or she hasn't encountered since medical school, computers don't need coffee and sleep and they don't forget information. Better yet, an AI system also may be able to track the latest medical research and even tap continuously into the observations and experiences of other doctors, and then crunch all that data to come up with statistically-validated treatment options [source: de la Torre].

AI researchers have been looking at ways to utilize such software in medicine for several decades [source: Patel, etal]. But the recent development of the artificial neural network, or ANN -- a program that utilizes the combined knowledge of its network connections and then learns from the data, in the fashion of a biological brain -- has really opened the door. In 2009, Mayo Clinic researchers reported that they had put data on 189 cardiac patients with implanted medical devices into such a system and that the software had screened them for potentially fatal heart infections with better than 99 percent accuracy. Not only did the software spot 72 of 73 implant-related infections, but it accomplished that without having to insert an endoscope down their throats to look around -- a potentially risky invasive procedure [source: Mayo Clinic].

While no one wants computers to replace human doctors, AI systems can help make them more effective in treating patients. Just how do these systems show promise for the medical field?


Computers as Digital Doctors

Machines don't have emotions -- at least not yet. So you have good reason to wonder about a digital doctor's bedside manner. But look at it this way: No matter how caring and compassionate that flesh-and-blood physician is, she may get things wrong an alarming amount of the time. Studies of autopsies, in fact, have shown that doctors seriously misdiagnose fatal illnesses in about one in five cases. And even more alarmingly, that error rate hasn't improved much since the 1930s, despite the vast amount of new medical knowledge and diagnostic tools -- ranging from MRI scanning to genetic tests -- that we've amassed since then [source: Leonhardt].

The unsettling fact is that human healers miss signs and make mistakes, and that's where computers can be of help. In the late 1990s, a British girl named Isabel Maude nearly died after her doctors diagnosed a flesh-eating bacteria infection as chicken pox. Her father Jason was so shaken by the experience that he decided to change his career. He left his job in corporate finance to found a new company, Isabel Healthcare, to help improve diagnoses. The company developed a program called Isabel, which allows doctors to type in a patient's symptoms. After probing a database of 100,000 medical sources, Isabel uses a set of specially developed problem-solving algorithms to generate a list of all the possible causes, including non-obvious ones that doctors may have overlooked. Within months of its launch in 2002, Isabel attracted 20,000 physician users in 100 countries. Today, the program is utilized in hospitals in the U.S. and abroad, and the American Medical Association offers it to members as a support tool on its Web site [source: Leonhardt,].


Recently, the National Library of Medicine has been working to develop a similar intelligent medical assistant that could answer patients' questions based in part on the answers to close to 200,000 queries posed by both medical and lay users at the Web site [source: PRWeb].

But the use of supercomputers, which can amass and crunch even more massive amounts of information, may enhance AI's potential even further.


Talking to an All-knowing Supercomputer Diagnostician

A doctor at Hackensack University Medical Center navigates the InTouch Health's RP-6 for Remote Presence robot, known as Mr. Rounder, while roaming the hallway in 2005.
A doctor at Hackensack University Medical Center navigates the InTouch Health's RP-6 for Remote Presence robot, known as Mr. Rounder, while roaming the hallway in 2005.
Spencer Platt/Getty Images

Isabel also has a brother of sorts, an IBM supercomputer called Watson. In 2011, Watson demonstrated its artificial intelligence prowess by handily defeating two flesh-and-blood champions on the TV game show "Jeopardy." But Watson has the potential to do a lot more than win trivia contests. Its developers envision using the system's ability to store and retrieve data to amass what could become the ultimate digital collection of medical data in existence. But that's not all.

Unlike less powerful systems, Watson has the ability to answer questions and analyze information in natural language -- that is, the way that humans express themselves -- and generate all the possible diagnoses that might come from the information. Not only that, but Watson actually can rank the diagnoses according to its understanding of the medical knowledge in textbooks, medical journals and reports of medical cases. In one demonstration of Watson's diagnostic abilities, researchers gave the system a fictitious case involving a patient who lived in Connecticut and had blurred vision and a family history of arthritis. Watson responded with a list of possible causes, topped by Lyme disease -- a seemingly unlikely diagnosis, but the one that researchers were looking for [source: Lucas-Fehm].


In the future, researchers envision adding speech recognition abilities to Watson, so that doctors will be able to communicate with their virtual advisor simply by talking into a smartphone or some other hand-held device. And information that doctors contribute to the system from their cases may be amassed and leveraged to help treat future patients. But there's also the possibility that medical supercomputers may expand their knowledge base beyond just research and doctors' observations. Conceivably, they also could begin data-mining lay sources, such as patients' blogs and comments made on medical Web sites. "What people say about their treatment ... it's not to be ignored, just because it's anecdotal," Dr. Herbert Chase, a Columbia medical school professor told Physicians News. "We certainly listen when our patients talk to us" [source: Lucas-Fehm]. And eventually, so will digital healers.

Author's Note

When I was growing up, I got frequent bronchial infections, and when my mom took me to the doctor, I was diagnosed with a long list of allergies to everything from pollen to chocolate to tomato sauce. As a result, I had to take daily doses of decongestants, and go back to the doctor's office every two weeks for immunotherapy injections. The shots didn't seem to do much good, and the decongestants gave me insomnia that left we walking around in a daze during the day. When I reached my mid-teens, I started running 3 or 4 miles every day and working out with weights, and as I got stronger and fitter, to my surprise, the symptoms gradually vanished. When I took another allergy test as an adult, nothing at all showed up. I don't know if I simply outgrew the allergies or my diagnosis was wrong. But I can't help but wonder whether an AI system might have come up with a more accurate diagnosis and/or a more effective treatment had such technology existed in the mid-1960s.

Related Articles


  • De la Torre, Christopher. "The AI Doctor Is Ready To See You." Singularity Hub. May 10, 2010. (Sep. 10, 2012)
  • Leonhardt, David. "Why Doctors So Often Get It Wrong." The New York Times. Feb. 22, 2006. (Sep. 10, 2012)
  • Lucas-Fehm, Dr. Lynn. "Watson: Extreme Evidence Based Medicine." Oct. 13, 2011. (Sept. 10, 2012)
  • Mayo Clinic. "Artificial Intelligence Helps Diagnose Cardiac Infections." Sept. 12, 2009. (Sept. 11, 2012)
  • Patel, Vimla L. etal. "The Coming of Age of Artificial Intelligence in Medicine." Artificial Intelligence in Medicine. May 2009. (Sept. 11, 2012)
  • PRWeb. "The National Library of Medicine Explores Artificial Intelligence Using Two-Hundred Thousand Real Patient Questions from" Jan. 5, 2012. (Sep. 11, 2012)