Path: utzoo!attcan!uunet!cs.utexas.edu!uwm.edu!rutgers!netnews.upenn.edu!msuinfo!sticklen
From: sticklen@cps.msu.edu (Jon Sticklen)
Newsgroups: comp.ai
Subject: Re: MODEL BASED DIAGNOSTIC EXPERT SYSTEM
Message-ID: <1990Jul4.002349.8729@msuinfo.cl.msu.edu>
Date: 4 Jul 90 00:23:49 GMT
References: <1357@ziggy.EDU>
Sender: news@msuinfo.cl.msu.edu
Organization: Michigan State University
Lines: 67

From article <1357@ziggy.EDU>, by cds@ziggy.EDU (Cindy Sarmiento):
> This is a call for help...
> 
> I am looking for advice or comments from anyone who has 
> experience in developing expert systems using model based reasoning
> or knowledge representation.
> 
> I am currently developing a medical diagnostic expert system and would
> like to use model based reasoning. I intend to use either an object-oriented
> ES shell like KAPPA from IntelliCorp or just an object-oriented language
> like C++, Smalltalk or Actor.
> 
> If you have experience or comments on this area, please respond by email.
> All replies will be gratefully received.
> 
> Thanks,
> 
> Cindy Sarmiento
> Univ. of South Florida (in Tampa, not Miami as many believe)

cindy,
  my mailer does not seem to know the name of your host - ie, my
e-mail got bumped. so i'll hope you see this answer.


cindy,
  there are a couple of places you might look. its a little
hard to know how to advise you without knowing more of
what kind of models you want to build. but...

casnet by kulikowski was one of the earliest approaches in medicine
that you might want to call model based. you can find descriptions of it
in almost any book on knowledge based systems.

ramesh patil did work in the early 80's on a system called
ABEL which was diagnostic in the area of acid-base-electrolyte
problems. it basically was a multi layered causal net which
he navigated via a set of defined "operators" which would
traverse the three fixed levels of his system. the major drawback
was that the reasoner required a starting place in the net
that was "reasonable". you can find a reasonable write up of
ABEL in the book "Readings in Medical AI" editted by Clancey and
Shortliffe.

another early (mid-80's) thing for you to look at is the CADUCEUS
system which was from pople and meyers. it also had the flavor
or a causal net, but it used the causal net in a wholly differnt
way as a kind of preprocessor for abductive problem solving.

i did a more recent work (late 80's) which included a diagnostic
compiled level system that interacted with a physiological model
represented as a "functional representation" of a thin slice of the
human body immune system (actually, the subsystem called the
complement system). a good write up of this work is in
Knowledge Acuqisition, Volume 1.

there is an exploding literature on model based reasoning, but very
little of it is in the medical area. you should look at the compilation
of papers on naive physics by  and deKleer. (i can't remember
the name of the book, but you should have no problem finding it with
deKleer's name.) the reading there include (i think) one paper by
kuipers who has applied hard core naive physics ideas in the medial area.
(his domain was rhenal failure if i remember right.)

good luck. its a very interesting area to be working it.

        ---jon---