Monday, June 27, 2011

The robot doctor will see you now - The Globe and Mail

Some doctors are known for their less-than-superb bedside manners. In their defence, they're only human.

But it's possible the human element will play less of a role in the future as machines take more responsibility for diagnosing diseases, assigning treatments and ensuring hospitals run smoothly and efficiently.

Don't expect computers to replace doctors. But as advancements in artificial intelligence continue to unfold, a growing number of computer experts, health professionals and businesses believe machines will have an increasingly important role to play. While that could result in more accurate diagnoses, fewer mistakes and cost savings, experts also warn that relying too heavily on machines could backfire. For instance, computers with inaccurate or incomplete information could give the wrong diagnosis, putting lives at risk.

Computer scientists who study artificial intelligence (AI) are developing machines that can provide answers to questions, identify key patterns in data and predict trends – to make machines that have human-like intelligence and abilities but can also operate more quickly and more efficiently, making fewer mistakes.

Medicine and health have emerged as areas that can benefit greatly. AI applications rely heavily on sifting through large amounts of data to identify patterns or come up with answers, which can make a difference in health care.

One important possibility is improved medical diagnoses. AI systems that are programmed to understand known diseases, symptoms and risk factors could quickly and easily make a diagnosis that a doctor could then verify in a physical exam. This means a patient could get access to vital medical treatments more quickly, which could be critical in certain situations. It could also save time and money.

"If you had somebody who had a really rare disease, then that might not be the first thing that the physician would think about," said Cory Butz, a computer science professor at the University of Regina, who spoke about AI recently at the University of Cambridge. "But the ... machine could go through all the possibilities and raise the issue, [saying] 'perhaps the person has this illness,' much quicker than the human would think about it."

The use of AI could also negate any bias that could affect a doctor's decisions. For instance, research has suggested some doctors have a bias against smokers, which could affect how that patient is diagnosed and treated.

The Mayo Clinic recently conducted an experiment to see whether AI systems could truly assist physicians. Researchers used "teachable software" that mimics the human brain to help diagnose potentially fatal cardiac infections in the hope of eliminating the need for an invasive exam.

The researchers introduced the system to countless real-world scenarios to help it evaluate symptoms correctly. After examining data from nearly 200 patients, the software correctly diagnosed the infection in nearly all cases.

The study was based on clinical data from previous patients, but if the software was introduced to hospitals, it could help patients and result in major savings. Similar systems could also help reduce medical errors, improve the accuracy of medical tests and help hospitals reduce inefficiencies.

Although it's clear many bugs need to be worked out before AI applications play a major role in day-to-day medical decisions, many are optimistic about future possibilities.

Recently, the famous IBM computer Watson appeared at a conference of health and computer science experts in Toronto. The computer was made famous earlier this year when it beat human contestants on the game showJeopardy! But company officials believe the computer's future lies in making medical diagnoses. Earlier this year, it announced a partnership with Massachusetts-based Nuance Communications Inc., a speech and imaging solutions firm, in a bid to get a product on the market.

But Prof. Butz also warned that any AI-based health care applications would need strict limits. Although machines are highly efficient, they are only as good as the information that has been programmed into them. For instance, the names of diseases could be fed into the machine incorrectly, or the system might not have the functionality to compare a person's symptoms against their risk factors for disease.

"I don't think that you would ever have a robot doctor or a robot physician," Prof. Butz said. "If you make a mistake diagnosing a patient, there could be drastic and even fatal consequences to that. The software system will always simply be aiding a human expert."

http://www.theglobeandmail.com/news/technology/the-robot-doctor-will-see-you-now/article2074992/?cmpid=nl-news1