ABSTRACT
This chapter explores the integration of two artificial intelligence approaches, temporal abstraction (TA) and Bayesian networks (BNs), in order to improve medical problem solving. BNs belong to the family of probabilistic graphical models, and they are widely used in many clinical domains as they can handle well uncertainty in medical knowledge and data. Moreover, TA methods aim to glean out the useful information/patterns from the clinical time-stamped multivariate data, in order to facilitate higher-level problem solving. The generated abstract concepts are divided into basic and complex TAs. BNs and TA demonstrated their effectiveness as stand-alone engines, predominantly for medical problem solving, but not in conjunction. This chapter investigates the claim that the integration of TA with BN could yield notable performance improvements in medical problem solving. Towards this end, we selected the field of coronary heart disease (CHD) as a test bed and demonstrator of the attempted integration, and we developed two models: (a) a temporal extension of a BN, namely, a dynamic BN, whose nodes represent basic TAs applied for prognosis of CHD, and (b) a naïve Bayes classifier whose features represent frequent temporal association rules, a type of complex TA, applied for the diagnosis of CHD.
