Integration of Temporal Abstraction and Dynamic Bayesian Networks in Clinical Systems. A preliminary approach

Abstraction of temporal data (TA) aims to abstract time-points into higher-level interval concepts and to detect significant trends in both low-level data and abstract concepts. TA methods are used for summarizing and interpreting clinical data. Dynamic Bayesian Networks (DBNs) are temporal probabilistic graphical models which can be used to represent knowledge about uncertain temporal relationships between events and state changes during time. In clinical systems, they were introduced to encode and use the domain knowledge acquired from human experts to perform decision support. A hypothesis that this study plans to investigate is whether temporal abstraction methods can be effectively integrated with DBNs in the context of medical decision-support systems. A preliminary approach is presented where a DBN model is constructed for prognosis of the risk for coronary artery disease (CAD) based on its risk factors and using as test bed a dataset that was collected after monitoring patients who had positive history of cardiovascular disease. The technical objectives of this study are to examine how DBNs will represent the abstracted data in order to construct the prognostic model and whether the retrieved rules from the model can be used for generating more complex abstractions.