Thursday, March 12, 2009

A thought about an application of computer vision

There is an article in the Technology Quarterly of The Economist (March 5, 2009) about recent advances in computer vision. For example, "a computer scientist at the University of Buffalo in New York, is designing software that helps determine the authenticity of expressions."
Fujitaka’s new machines refuse to sell cigarettes if their software detects plumpness in the skin (a tell-tale sign of adolescence) around a potential customer’s eyes. Tests show that the system is slightly better at estimating people’s ages than nightclub bouncers are.......

Computer scientists at the Toronto Rehabilitation Institute in Canada have been testing a computer-vision system for monitoring people living in nursing homes or alone. A cheap camera, stuck to the ceiling, wirelessly relays images to a small computer that monitors how people move.
I wonder whether computer vision systems could be developed to help in the phyzician's consulting room? The examples above suggest potentially useful applications, such as detecting when a patient's responses are not authentic, or estimating the age of a patient. Indeed, monitoring movement might provide insights for the physician.

It is my impression that in the past doctors got a lot of information from the appearance and even the smell of their patients. It might be that the current reliance on computer imagry and laboratory tests has reduced not only their reliance on the appearance of patients but even their attention to that appearance.

It might be that appropriately developed computer vision systems could help physicians to attend to and interpret patient appearance, and even that with super computers and improved sensors, more diagnostic information could be inferred from appearances than even the best old time practitioners would have achieved.

4 comments:

Anonymous said...

Nice article, however all those vision systems will unfortunately have little percentage of errors and in case of the vision systems used by doctors even those small errors could cause terrible results. Even if one person in a thousand died because of the error of the vision system it still would be a big loss. I still do agree that vision systems could be used widely but just in the situations where human life is at stake.

Take care, Elli

John Daly said...

Thank you for the comment.

Most of the people who show up in a doctors office are either well and just worried or a suffering from a health problem that is self limiting.

Still, my idea was that a computer vision system might help a doctor to do a better job of diagnosis and communication, leaving the decisions to the human physician.

I do get your point. I once worked in a research lab where another group was developing a voice warning system for pilots. The first time it was tested, the pilot of the plane landed with the landing gear up and destroyed an expensive plane. When asked why he did such a thing he responded "I was trying to figure out what was wrong but that darned voice warning system in my ear kept distracting me."

You have to be very careful that a system designed to help people make better decisions doesn't backfire and result in worse ones.

B Grey said...

You could make a similar argument about using computer vision to check for say, breast cancer. There will be a risk of missing something that should have been detected. But on the other hand, a computer is more consistent than a human: on a good day a human can be a very good inspector, but they get tired, bored, distracted, and so can miss the obvious.

The challenge for the vision engineer is to (a) develop a system that errs on the side of caution - in other words, gives false positives but not false negatives (do I have that the right way round?) and (b) signals that human review is required, rather than making a decision.

B Grey,
http://machinevision4users@blogspot.com

John Daly said...

And in fact, I recall studies of the delegation of medical tasks which indicated that under some conditions non-physicians make fewer medical errors than physicians.

The explanation is that someone trained to do a narrow range of things, doing the same things over and over again, can be more accurate that a very highly trained generalist who does a great variety of things, and some of those things very infrequently.