Applying Object Oriented Bayesian Networks to Large Medical Decision Support Systems

This paper describes the use of the object oriented Bayesian network framework in two applications in the medical domain. The first example models the glucose metabolism in humans and is intended for planning of insulin injections for diabetics. The main characteristic of this application is a temporal repetition of identical model structures, where the basic building block consists of a one hour model of the metabolism. This type of model is usually modeled as a dynamic Bayesian network, and we show how object reuse, and in particular the concept of time slices, can be exploited in the construction of such models. The other application is the MUNIN system for diagnosis of peripheral muscle and nerve diseases, that is characterized by a number of (almost) identical anatomical structures. The modeling of such structures benefit from inheritance properties of object oriented Bayesian networks, and we further illustrate how time slices can be used to combine partial contributions of some effect to an overall description of the joint effect of those contributions. It is concluded that the virtues of the object oriented framework eases the specification and maintenance of large decision support systems.