Design of a case-based reasoning system applied to neuropathy diagnosis

Abstract

The goal of our work is to develop a computer assisted medical diagnosis system in the field of neuropathy (peripheral nervous system diseases). We believe that an efficient medical diagnosis system must integrate the divers reasoning capacities of physicians: logical, deductive, uncertain, and analogical reasoning. A neuropathy diagnosis system based on production rules, called Neurop, has already been developed in our laboratory. However, Neurop is poorly adapted to treat uncertain data or unusual cases.

In this paper, we propose integration of a new module of analogical reasoning to compensate the drawbacks presented by Neurop. The idea is to memorise real treated cases and to construct what we call a memory of prototype cases. This memory is then used during retrieval phase. A learning phase is added to optimise the system reaction and to prevent future diagnostic failures. This is achieved by modifying the contents of the prototype memory. Three principal types of modifications are offered. They are: prototype construction, prototype specialisation and prototype fusion. The proposed reasoning system is planned to function in conjunction with another reasoning system (that could be a human expert) which supervises results and activates the learning mechanisms in case of failure.