Experience with a Model of Sequential Diagnosis

Abstract

In recent years, a number of studies of the use of computer programs in diagnosis have been performed.Central to each of these efforts has been the development of an explicit, precisely formulated procedure for diagnosis.Such a development is a prerequisite for computer programs of this type.In general, attention has been focused on models of the inference function of diagnosis, the development of a diagnosis from the given set of clinical signs.Some interesting probabilistic models have been developedwhich employ Bayes rule, cB ayes rule has understandable appeal for use in such a model.First, it permits the use of probabilities in inference, This is preferable to a deterministic approach, because it reflects some of the basic uncertainties of diagnosis.Also, Bayes rule provides a rational means for considering both a_ priori belief about the incidence of various diseases and the evidence embodied in the clinical signs in a given case.Finally, the formulation of the inference function in terms of Bayes rule is particularly suited for incorporation into a computer program.Given the necessary statistical data, the problem of inference is thereby reduced to a problem of computation.While the Bayesian model is well suited for computer diagnosis, there are certain problems associated with its use.First, the model requires that extensive statistical data be available for the given area.The