|title:||Modeling physiological processes with dynamic Bayesian networks|
|published in:||September 2006|
Master of Science thesis
Man-machine interaction group
Delft University of Technology
|PDF (9794 KB)|
Traditionally, the medical domain has been reluctant to introduce information and communications technology (ICT) in the form of decision support systems or patient databases on a large scale, being afraid that the craft of healing will get mechanized. However, the tide is changing as more and more physicians see the benefits that ICT can provide them, and understand that machines will never replace a human's judgment. The research presented in this thesis provides a framework on how to apply DBNs for modeling physiological processes in the human body. During our research, the existing DBN formalism proved to restrictive and that is why we choose to extend it. Our extension includes support for kth-order temporal arcs, and introduces the concepts of a temporal plate, and anchor, contemporal and terminal variables. The extended DBN formalism was implemented and validated in SMILE and GeNIe, the two reasoning software solutions developed at the Decision Systems Laboratory of the University of Pittsburgh. The implementation enabled us to use this software for temporal reasoning purposes, and is to the best of our knowledge the first implementation of temporal reasoning that provides support for kth-order temporal arcs, a temporal plate, and anchor, contemporal and terminal variables, next to support for using different canonical representations in temporal models, such as the Noisy-MAX gate. Furthermore, we present our methodology for obtaining the DBN structure and learning the parameters for our extension of the DBN formalism, including issues such as the discretization of continuous variables. Finally, we apply our extended DBN formalism in collaboration with Philips Research to two physiological processes: the glucose-insulin regulation and the cardiovascular system. Empirical studies show that our extension of the DBN formalism provides an intuitive and sufficient way for modeling physiological processes with a DBN, but many of our findings can be easily generalized for the modeling of other domains.