Diagnostic Reasoning from the viewpoint of Rough Sets

Abstract

In existing studies, diagnostic reasoning has been modeled as if-then rules in the literature. However, closer examinations suggests that medical diagnostic reasoning should consist of multiple strategies, in which one of the most important characteristics is that domain experts change the granularity of rules in a flexible way. First, medical experts use the coarsest information granules (as rules) to select the foci. For example, if the headache of a patient comes from vascular pain, we do not have to examine the possibility of muscle pain. Next, medical experts switches the finer granules to select the candidates. After several steps, they reach the final diagnosis by using the finest granules for this diagnostic reasoning. In this way, the coarseness or fineness of information granules play a crucial role in the reasoning steps. In this paper, we focus on the characteristics of this medical reasoning from the viewpoint of granular computing and formulate the strategy of switching the information granules. Furthermore, using the proposed model, we introduce an algorithm which induces if-then rules with a given level of granularity.