1. I accept that I am under the control of a higher power (Muppet).

Differential diagnosis web site?

I'm amazed that on House MD the doctors just gather 'round the whiteboard list of symptoms and try to come up with diseases.  Don't they have expert systems that do this for you?  Do doctors really try to memorize every single disease on record?

And now for the real question: is there a web site that lets you key in symptoms and it'll do the differential diagnosis for you?  That'd be so cool.
Permalink Michael B 
March 21st, 2007 9:17pm
In an old AI course the teacher talked about a medical diagnostic program that used relatively simple forward and backward chaining techniques. After years of development it did a pretty good job. I never understood why it wasn't used, except perhaps that most problems are relatively simple. But was just reading a paper that said up to 25% of diagnosis were wrong, so I don't think do that good a job.

The missing link I think is our primitive diagnostic incapabilities. They do one test and then another. Why not get some blood and test everything. That would feed into the network.
Permalink son of parnas 
March 21st, 2007 11:21pm
Diagnostic trees are practically their own AI.

It's the obscure/cutting edge stuff that doesn't get into the standard medical diagnostic texts that's hard to diagnose.
Permalink Send private email ~~~x 
March 21st, 2007 11:38pm
> s hard to diagnose.

Up to 25% of diagnosis are wrong. So it must be a little bit harder than that.

http://management.curiouscatblog.net/2007/03/19/errors-in-thinking/
Permalink son of parnas 
March 21st, 2007 11:41pm
the trouble with differential diagnosis is that there are so many illnesses that all have exactly the same symptoms.

in most cases - especially the common ones - you end up with a sodding great list of possibilities.

its only once you start looking at the specific context of the patient that you can begin to derive an accurate one, if you watch closely that is a large part of what house does.

they do exist though, and hospitals are increasingly using them.
Permalink Send private email zestyZucchini 
March 22nd, 2007 4:15am
>In an old AI course the teacher talked about a medical diagnostic program that used relatively simple forward and backward chaining techniques. After years of development it did a pretty good job. I never understood why it wasn't used, except perhaps that most problems are relatively simple.

That would most likely be MYCIN.
http://en.wikipedia.org/wiki/MYCIN

Expert physicians would have a correct diagnosis rate of about 80%.
MYCIN had a correct diagnosis rate of 65%.
Non-expert physicians had a correct diagnosis rate near 40%.

Folks were more worried about liability, so they never released it.

http://en.wikipedia.org/wiki/CLIPS
If you ever used CLIPS (or jess), it will look very familiar, with rules like:

  (defrule 52                                       
      if  (site  culture        is  blood )
          (gram  organism        is  neg  )
          (morphl organism        is  rod  )
          (burn  patient        is  serious)
      then 0.4                                       
          (identity  organism    is  pseudomonas )
  )

which sort of translates into:
rule_52:
IF the culture came from blood,
AND the organism is gram-negative
AND the organism is rod shaped
AND the patient has serious burns
THEN it is 40% likely that the organism is pseudomonas.

http://www.cs.columbia.edu/~laza/Software/Mycin/mycin.html
Permalink Peter 
March 22nd, 2007 6:02pm

This topic is archived. No further replies will be accepted.

Other topics: March, 2007 Other topics: March, 2007 Recent topics Recent topics