Slashdot Mirror
Dogs Sniff Out Bladder Cancer From Urine
hookedup writes "Volunteers from Hearing Dogs for Deaf People found dogs can be trained to sniff out a tell-tale scent of bladder cancer from urine. As a group, the 6 dogs used in the test correctly flagged the positive sample 22 times out of 54, for a success rate of 41 per cent. By chance alone, the expected success rate would be 14 per cent"
If something has even the faintest smell, it's almost certain that dogs can smell it. I think the main trouble is describing to the dogs exactly what you want the dogs to find and getting good ways of letting the dogs tell you.
There are dogs who can tell when people are about to have an epileptic fit or seizure, sometimes even 15 minutes before. They'll give a signal so that the people can go find a nice safe place to lie down or whatever, and prepare themselves for the seizure. Dogs can also find land mines based from air samples taken from various zones and put into test tubes. Dogs can also find bodies under many feet of _flowing_ river water. They have been trained to find skin, lung and other cancers.
I think it's mainly a communication problem (plus some dogs also do get bored, and a relationship thingy - you're not dealing with machines).
It's like having someone born without a sense of smell trying to get someone with a sense of smell to say whether a food is prepared using wine or not. And both don't speak the same language. All you can say to the person is "yes" you got it right. And "no" you don't. And this is based on the results of a machine (made by people without a sense of smell) which is not necessarily 100% reliable either. There's plenty of scope for misunderstanding. For example: the person could think you're trying to identify food with _alcohol_. Or food with grape juice. Or maybe it turns out that the samples have some other correlation (e.g. most of the positive test samples had people who ate/drank something the same within 24 hours before they peed).
Note also that the dogs found someone with kidney cancer. So the dogs could also be right that it does smell like cancer - the people may indeed have cancers or precursors to cancers _elsewhere_ in the body. If the samples are from older folk, that might be true. Many people probably have cancers somewhere, it's just their immune systems keep them suppressed or had a lapse and are now busy getting rid of them.
And not all the dogs have similar skills nor have similar understanding of what you want. So a star performer who gets it, would probably do orders of magnitudes better.
They're trying to make a machine that does it. Maybe it won't be that difficult. But it could be hard if it's like detecting something _qualitative_ and not _quantitative_. Example: you're trying to find a sample that smells _mainly_ like a particular person but has traces of something that smells like that person just slightly different in a "bad" way. e.g. a sample that smells like Bob plus a faint smell like Bob going bad (whatever "bad" means).
e.g. it's like detecting which sheets of paper have one or two off-colour spots on them. And all the sheets are differently coloured. If you only have a machine that only detects the average colours you don't have a chance.