Rich Neapolitan a �crit:

>Dear Colleagues,
>
>I would like to know of any other successful medical expert systems. By 
>successful I mean one that is actually being used to assist 
>physicians/patients in making decisions. It need not be making money.
>
>Thanks,
>Rich
>
Until November 26, I was a PhD student (now I'm a Doctor ;-) ) and I was 
working on dynamic bayesian networks to monitor hydration rate of kidney 
disease patients. The Diatelic project at INRIA Lorraine 
(http://www.loria.fr and http://www.diatelic.com (sorry it's in french)) 
is a long-term telemedicine project (1997) to make medical monitoring 
systems aiming at computing the hydration rate of kidney disease 
patients doing a dialysy treatment. The patients stays at home and send 
every day a set of physiological data (weight, temperature, blood 
pressure, and others kidney specific data) to a remote server which 
hosts the monitoring system.
The current commercial version (v2) is using  a POMDP and monitor 15 
patients since 2 years. Results are : regulation of blood pressure and 
weight of patient (which are very unstable for kidney disease patient), 
regulation of the hydration rate (which mainly depends on blood pressure 
and weight), drop of average medical costs. Medical and economic results 
are good !
During my PhD thesis, I have worked on a new Dynamic Bayesian Networks 
version (v3) which is able to deliver a more complete and accurate 
diagnosis to the physician by using more data and information than the 
previous version (which is quite simple but works fine). The system 
delivers a diagnosis on the hydration rate and a diagnosis on the 
viability of an ideal weigth. Others information are also avalaible with 
the new DBN system.
Some results given by the pomdp system was wrong sometimes but this is 
no more the case with the DBN system. The physician is now able to take 
a daily therapeutic decision by using the diagnosis given by the DBN. 
The most exciting part of such a problem is that the hydration rate and 
related information are hidden and cannot be efficiently and daily 
measured, and the thruth about the real diagnosis is quite impossible to 
know for a given patient (this is often the problem in medical diagnosis 
and monitoring). Others problems are : learning the particularity of 
each patient, estimating the quality of the diagnosis and adapting 
parameters of the system given the recommendations of the physician, etc...

Since Novembre 27 (!), I'm mainly involved in dynamic bayesian networks 
(of course ;-) ), variationnal methods. I'm also involved in all that 
stuff which are very exciting about bayesian networks : learning, 
adaptation and various applications (with other people) like medical 
diagnosis, robotics, information retrieval,...

If you would like to know more about the Diatelic project, I could send 
you papers (english) and/or my PhD thesis (french) about DBN and Diatelic.

Best Regards,
David Bellot
[EMAIL PROTECTED]
INRIA Grenoble, France

Reply via email to