Preconditioning of Measurement Data for the Elimination of Patient-Specific Behavior in Qualitative Modeling of Medical Systems

Keywords

Abstract

A major problem in the biomedical domain is knowledge generalization. Is knowledge acquired from and about one patient at all applicable to another, and if so, to what extent? Can an inductive qualitative model acquired by analyzing data retrieved from one patient be used to predict the behavior of another? The purpose of this paper is to enlighten all this questions by means of a qualitative methodology called fuzzy inductive reasoning.

To this end, a technique based on the combination of different patients knowledge is presented in this paper that enables to obtain a model for a specific class of similar patients/operations pairs. A new feature called missing data option is used to allow the creation of fake causal relationsships. This feature is essential in order to be able to combine several data sets.

A medical application based in the control of anaesthetic agent to be applied to a patient undergo a clinical operation is used to demonstrate the feasibility of this method. Two data sets from different patients undergoing different operations have been used in order to obtain a unique model that identify a similar patient/operation class.

It will be shown that the predictions obtained by this common model are not as good as those obtained from each patient with its specific models. This is reasonable and charactesristic of all knowledge generalization processes. Those results are still significant and useful for medical advice in the operation theater.

Interested in reading the full paper? (6 pages, 146,280 bytes, pdf)

Homepage

Last modified: June 17, 2005 -- © François Cellier