Approximate reasoning and interpretation of laboratory tests in medical diagnostics

In many real world domains, such as medicine, human knowledge is by nature imprecise. As a consequence the expert systems oriented to these domains must have specific tools to deal with the uncertainty. A commonly used approach is to equip the expert system with a computational capability to analyze the transmission of uncertainty from the premises to the conclusion. The theory of fuzzy sets provides a systematic framework for dealing with fuzzy quantifiers, such as many, few, most, increased, normal, etc. In this framework the results of laboratory tests (we refer to the area of medical diagnostics) which are precise may be interpreted in the terms of fuzzy propositions by means of membership functions. In this paper we describe our approach to the construction of the membership functions using the statistical data. We show how to build a membership curve for a fuzzy set of 'normal' values using a probability distribution and a given range of values considered as 'normal', and then we describe a method of interactive refinement of such a curve.