International Journal of Hybrid Intelligent Systems
Volume 2, No. 1(2005), pp. 35 - 55
A Hybrid Model for Diagnosing Multiple Disorders
Sabina Munteanu
Abstract:
This paper presents a two-leveled hybrid system for medical diagnosis. The first
level is responsible for hypotheses’
selection and uses an original association-based reasoning scheme (shallow
knowledge), which measures the distance between
the observations and the prototypical models of diseases, using fuzzy decision
functions. This first module is efficient but
not very precise and hardly transparent; the key of the representation it uses
is to understand symptoms which occur within a
disease’s definition as fuzzy criteria, and to aggregate these criteria in a
unique complex decision function which models the
disease as a whole. The reduced problem made up of the hypotheses selected in
the first step is passed through a refining
and discriminating process, which was inspired from direct argumentation systems
(the latter being a relatively new approach
to diagnosis problems). This part is meant to provide explanation facilities for
the results and to remove the contradictions
generated by the previous step, if any. Original definitions are suggested for
argument (a monotonic structure replacing
abductive explanation) and attack (representing the defeasibility/non-monotonicity
of reasoning). It is more efficient to reduce
non-monotonicity to an attack relation between arguments, than any of the
approaches used by hypothetical-deductive abduction.
A brief theoretical analysis of the model within Ginsberg’s unified framework of
multivalued logic is finally presented.
Keywords: Diagnosis, hybrid system, fuzzy decision function, causal network, direct argumentation system, multivalued logic
Copyright © 2004 Advanced Knowledge International, Australia