Methodology for a medical expert system on fuzzy analog ganglionar lattices. Non-approximate reasoning with multiple antecendents of different relative importance and limited uncertainty.

Abstract

This work presents an Expert System based on fuzzy analog ganglionar lattices. Its reasoning scheme is designed analogously to the expert's mental organization and it is realized on an (analog) operator called the ganglionar lattice. It is a connectionist system that uses the medical knowledge to define its architecture. The operator evokes some similarities to higher order neural networks and performs as the knowledge base and inference engine of the expert system, in a unified manner. A main feature of this operator is that it exhibits the variables corresponding to all intermediate concepts identified by the expert; this characteristic is shown to be most valuable for assessing, explicating and prospecting in medical applications. Further, it is capable of (i) evaluating a consequent for a variety of non-approximate reasonings with multiple antecendents of different relative importance under limited uncertainty; (ii) explicating the conclusions at different levels of abstraction to suit the user; and (iii) prospecting for the best 'a priori' sequence of unevaluated antecedents, from which to choose following tests. These procedures are based on the objective criterion of the consequent's uncertainty decrease (entropy). All results are produced in numerical form and may be translated into restricted natural language. A simple example of this technology is fully developed. Finally the method's potentials are discussed for future applications.