A new hybrid case-based architecture for medical diagnosis

Abstract

This paper proposes a new hybrid case-based architecture, which supports multiple-disease diagnosis and the learning of new adaptation knowledge. The architecture combines case-based reasoning (CBR), neural networks, fuzzy theory, induction, utility theory, and knowledge-based planning technology together to facilitate medical diagnosis. The basic mechanism is that of CBR. A distributed fuzzy neural network is employed to perform approximate matching and thus to tolerate potential noise in case retrieval. The induction technology along with utility theory is used to select valuable features of the target case and prune unnecessary search space. Knowledge-based planning is a general-purpose mechanism for case adaptation. It creates a case adaptation plan from an adaptation tree, which contains all relevant problem features, satisfies all relevant constraints, and contains all cases whose expected utilities are greater than a threshold. Execution of the case adaptation plan leads to the diagnosis of multiple diseases. The adaptation tree facilitates the reuse of cases and the learning of various types of knowledge-including relationships between disease types and features, case-specific verification knowledge, and differential diagnosis rules. Integrating these techniques in the CBR paradigm can effectively produce a high-quality diagnosis for a given medical consultation.