Human-like reasoning capability in a medical diagnostic system: The application of fuzzy set theory to computerized diagnosis
The rationale, theory and outline of a medical diagnostic system based on fuzzy set theory have been presented. The hope is that such a system, when fully developed, will be able to simulate non-quantitative human reasoning. Preliminary computer runs have shown that the fuzzy goodness of fit approach using a simple algorithm may produce reasonable results when given data on patients with acute renal failure. Work is underway to implement this system in an interactive mode and to inject artificial intelligence algorithms as described earlier. In order to provide explanations of the logic for users, procedures similar to the ones used by Shortliffe et al. will be implemented. This will enable users to ask 'how' and 'why' when the computer prints out diagnoses or asks for data. We feel this system, with its probabilistic, deterministic and fuzzy goodness of fit capabilities is a versatile one, able to compromise between the mathematically precise but clumsy Bayesian systems and the earlier network systems with their inability to store a great deal of useful data.
