Dear Rich, We are developing since 1998 medical diagnostic systems for internal medicine using the Bayesian networks.
The maturity of our work can be described as 'close to clinical practice'. We have just finished the first clinical trials to evaluate the acceptance of a module on lipids and vascular diseases (approx 500 variables) by the intended users, 15 experts in internal medicine working in a hospital environment. The response is overwhelmingly positive. The objective of our project is to build a large bayesian network for diagnosis in internal medicine. As is well-known, this is not easy and requires the combined efforts of experts in internal medicine as well as advanced software and algorithmic development. On the medical side, one of the distinguishing features of our work is that we have the financial resources to contract physicians to do the medical modeling and evaluation in a clinical setting. In our view, this is critical to 1) obtain valid models and 2) to get accepted by potential users. On the algoritmic side, our research group in Nijmegen has experience on approximate inference techniques that are needed to keep computation tractable. We have developed our own graphical model software, called BayesBuilder, which is freely available for non-commercial use. We recently completed a report describing the state-of-the-art of this project: "PROMEDAS": a probabilistic decision support system for medical diagnosis The report can be downloaded from http://www.snn.kun.nl/~bert/#diagnosis As an aside, we have used the same technology to build an diagnostic expert system for the diagnosis of bearing malfunction for SKF, which is one of the worlds largest bearing manufacturers. The system contains 90 variables and is operational as SKF since 2001. The tool is accesible to all SKF employees world wide through an intranet/servlet application. I hope this is of your interest. Best regards, Bert Kappen SNN University of Nijmegen tel: +31 24 3614241 fax: +31 24 3541435 URL: www.snn.kun.nl/~bert
