Applications of probabilistic and related logics to decision support in medicine

Abstract

Since the late 60s, probability theory has found application in development of various medical expert systems. Bayesian analysis, which is essentially an optimal path finding through a graph called Bayesian network, has been (and still is) successfully applied in so-called sequential diagnostics, when the large amount of reliable relevant data is available. The graph (network) represents our knowledge about connections between studied medical entities (symptoms, signs, diseases); the Bayes formula is applied in order to find the path (connection) with maximal conditional probability. Moreover, a priori and conditional probabilities were used to define a number of measures designed specifically to handle uncertainty, vague notions, and imprecise knowledge. Some of those measures were implemented in MYCIN in the early 70s [96]. The success of MYCIN has initiated construction of rule-based expert systems in various fields.