An ontology for model-based reasoning in physiological domains
Nuri Serdar Uckun · 1992
A recent focus of artificial intelligence in medicine is on qualitative modeling and reasoning. Qualitative reasoning methodologies are inherently tolerant to missing and inadequate information. These methods are useful in domains where more precise models cannot be derived, or for applications where the use of detailed models may not be necessary. However, unconstrained use of qualitative models in large problem domains results in excessive ambiguity. This dissertation describes the design, development, and application of a hybrid ontology for modeling and reasoning in physiologic domains which combines the advantages of qualitative and numerical methods. The ontology, named Y scAQ, is based on the qualitative simulation principles put forth in the Qualitative Process theory (QPT). However, Y scAQ is based on a hybrid algebra of qualitative and numerical constraint equations. Y scAQ integrates capabilities for interpretation, model-based diagnosis, and prediction in a data-rich environment. The ontology defines a generic language in which physiologic models may be developed, revised, and reused with relative ease. Y scAQ is optimized to operate within a distributed intelligent monitoring architecture where it is responsible for maintaining a high-level interpretation of patient status and providing a context mechanism to regulate the activities of other modules. The intelligent monitoring architecture is applied to the problems of monitoring and ventilator management on newborns with respiratory distress syndrome (RDS). The interpretive and predictive capabilities of Y scAQ are evaluated on a model of RDS and assisted ventilation. The evaluation is based on actual clinical data obtained from patient records, and on data derived from experiments on newborn lambs. The results and the limitations are discussed and future research directions are outlined.