An expert system for the analysis and interpretation of evoked potentials based on fuzzy classification: application to brainstem auditory evoked potentials.

EPEXS is an expert system for evoked potential analysis and interpretation (a medical examination performed in clinical neurophysiology laboratories), working from available clinical records and numerical data extracted from evoked potential traces. EPEXS integrates two formalisms of knowledge representation: rules and structured objects. The rules represent the elementary concepts (shallow knowledge) and include a model of possibility based on the Dubois and Prade default reasoning and possibility theory. The structured objects (prototypes) are organized as hierarchical taxonomies (underlying knowledge). These allow the description of both the objects and their relationships. The heuristics used to interpret knowledge are based on two hypotheses: the unicity of the pathological process leading to several given symptoms and the progression from the general to the specific, leading to the adoption or rejection of a class of diagnoses. This avoids the problem of the differential diagnosis. These sources of knowledge are used in a dynamical way that could be described as a four-step process: acquisition of clinical data in order to define the nosological frame of the pathology, production of hypotheses about the nature and topography of lesions, interpretation of data in accordance with these hypotheses, and finally evaluation of their likelihood. The validation shows that EPEXS topographic diagnoses were correct in 100% of cases and 92% of it nosologic diagnoses were correct, and no pathological record was interpreted as normal. When examined on a given pathology basis EPEXS was not significantly different from human experts as regards to performance, specificity, and sensitivity.