November 11, 2006
Google Searches Help Doctors Diagnose Cases

Stumped by a difficult patient and don't have a real life Gregory House MD to turn to? Google it!

Searching with Google may help doctors to diagnose difficult cases, finds a study from Australia published on bmj.com today.

Doctors have been estimated to carry two million facts in their heads to help them diagnose illness, but with medical knowledge expanding rapidly, even this may not be enough. Google is the most popular search engine on the world wide web, giving users quick access to more than three billion medical articles.

This is where cyborg technology would come in handy. Think a question. Get a search result projected onto your retinas. Or even state a question out loud. Then have a device interpret your speech into text and send it off to a specialized search engine. Then see results.

Just as Google has build Google News (of which FuturePundit is an incredibly heavy user) to restrict searches to news sites and Google Scholar to restrict searches to scholarly published research papers imagine a Google Medical that only looks at medical sites (and avoids all those places trying to sell you supplements). That would provide even more useful results than plain Google.

Google searches allowed doctors to turn up correct diagnoses in 58% of 26 difficult cases.

So, how good is Google in helping doctors diagnose difficult cases?

Doctors at the Princess Alexandra Hospital in Brisbane identified 26 difficult diagnostic cases published in the New England Journal of Medicine in 2005. They included conditions such as Cushing’s syndrome and Creutzfeldt-Jakob disease.

They selected three to five search terms from each case and did a Google search while blind to the correct diagnoses.

They then selected and recorded the three diagnoses that were ranked most prominently and seemed to fit the symptoms and signs, and compared the results with the correct diagnoses as published in the journal.

Google searches found the correct diagnosis in 15 (58%) of cases.

But you need to have a large base of knowledge in your own brain already in order to make sense of the search results.

The authors suggest that Google is likely to be a useful aid for conditions with unique symptoms and signs that can easily be used as search terms.

However, they stress that the efficiency of the search and the usefulness of the retrieved information depend on the searchers’ knowledge base.

Medicine is the area of intellectual I'd most like to see automated. Medical costs go up each year faster than inflation and have become a huge burden for industries, governments, and individuals. The vast majority doctors and nurses employed in medicine are smart minds that only deliver services and do little to develop new goods and services (though obviously some do research). The automation of medicine would free up hundreds of thousands of higher IQ minds to do research and product development.

Artificial intelligence is going to come very incrementally. The popularity of search engines is probably a bigger driver for the development of artificial intelligence than anything else happening right now. Hundreds of millions of people are using search engines looking for answers. The advertising revenue they generate feeds an intense competition between search engine providers to make better algorithms to search through text and tables to look for meaning.

Even the people who do searches to look for the latest in the trailer trash saga of Britney Spears and Kevin Federline are creating demand for products that push the boundaries of artificial intelligence. The people who want to find others who share their own weird quirky sexual desires are going to jump to the search engine that can interpret human writing most accurately. So all the vices, cravings, and desires of humanity are pushing us further toward development of useful artificial intelligence technology. Parenthetically, this also suggests that some of the first AIs will, by design, want to have sex with humans. So Caprica-Six's desire for Gaius Baltar makes sense.

Share |      Randall Parker, 2006 November 11 09:41 AM  AI Medicine


Comments
rsilvetz said at November 11, 2006 10:38 AM:

Not quite AI, but back in the day, Dr. Larry Weed convincingly showed how to improve medical practice with computers using an integration of theory and practice called knowledge coupling".

Interestingly, just plain old recursive intersection of sets increases diagnostic accuracy. Some of Dr. Weed's work showed you can push north of 90% pretty quickly (relative to pencil and paper).

I for one welcome our future AI friends, when can I order my personal Caprica-Six model?

James Bowery said at November 11, 2006 1:00 PM:

Eventually, and it doesn't look too far in the future, there will be public funding for competitions such as the Hutter Prize for Lossless Compression of Human Knowledge which will attack the AI problem at its root rather than nibbling around the edges with Britney Spears queries.

The importance of this prize is starting to make headway and in time will unleash an epistemological revolution focused on quality control of the explosion of information. That will have more profound implications than the revolution of quality control in manufacturing.

gcochran said at November 11, 2006 8:38 PM:

There have been useful Ai program for medical diagnosis for a long time. They can usually beat anybody but a committe of nationally-ranked experts at diagnosis and will run on a ten-year old PC. Naturally, they are not much used.

James Bowery said at November 12, 2006 11:06 AM:

If there are any such proven high-accuracy diagnostic systems in the public domain or under an open source license (which one would hope to be the case if they've been around that long) I'd like to take a shot at putting one on-line with a web interface.

Certainly computation load shouldn't be much of a problem given how much computation power has progressed over the intervening period.

Gerald Hibbs said at November 13, 2006 2:43 AM:

"If there are any such proven high-accuracy diagnostic systems in the public domain or under an open source license (which one would hope to be the case if they've been around that long) I'd like to take a shot at putting one on-line with a web interface."

A little late on this one, here is at least one diagnostic AI (of a sort):

http://www.wrongdiagnosis.com/symptomcenter.htm

You simply click on the problem area and then scroll down a bit to add more symptoms. WebMD, as well as other sites, offer similar tools.

Here is another that is specifically designed to diagnose from symptoms:

http://www.myelectronicmd.com/step2.php
"Find your diagnosis by analizing (sic) your medical symptoms by using our free online health screening engine."

This type of thing is fairly trivial to do. Also note that the wrongdiagnosis.com website also serves up this article prominently:

"http://www.wrongdiagnosis.com/diagnosis/self.htm
Self diagnosis is a dangerous practice. In fact, it is one of the most likely ways to get a misdiagnosis, which is the one thing we want to avoid. We recommend you always seek prompt professional in-person medical advice from your local qualified medical professional."

Ok, here is a last link, I'm sure there are tons of others if I looked about a bit more. They charge a small fee, about $15.00:
http://www.yourdiagnosis.com/
"YourDiagnosis asks questions about every body system. No symptom will go undetected.

Your allergies, medications and immunizations are recorded. Also, your family history and past medical problems.

YourDiagnosis does a complex analysis of all information gathered about your symptoms. It then produces a list of all possible medical diagnoses matching your symptoms."

James Bowery said at November 13, 2006 9:11 AM:

It's one thing to come up with a diagnostic website. Anyone can do that. It's another to come up with a diagnostic website that "can usually beat anybody but a committe of nationally-ranked experts" as asserted by gcochran.

Cites would be appreciated.

Post a comment
Comments:
Name (not anon or anonymous):
Email Address:
URL:
Remember info?

                       
Go Read More Posts On FuturePundit
Site Traffic Info
The contents of this site are copyright ©