Application of fuzzy reasoning in an expert system for ultrasonography.

Abstract

To apply fuzzy reasoning in a computer-assisted diagnostic system for ultrasonography and to evaluate its clinical utility as a diagnostic aid for the unskilled clinician. A computer-assisted diagnostic system for ultrasonography was constructed to differentiate metastatic from inflammatory lymph nodes. The pattern of a skilled clinician's considerations was simulated with the fuzzy expert system. Three fuzzy production rules were set up according to the diagnostic criteria for lymphadenopathy. Each rule was transformed into a membership function. The max-min composition method was used for inference and the centroid method for defuzzification. The system was tested with 20 clinicians who were one to three years after graduation and inexperienced in ultrasonography. Accuracy, sensitivity and specificity were evaluated. The average increase in accuracy was 8.5%, and the sensitivity and specificity 10.7% and 6.4% respectively. These scores were significant for accuracy (P = 0.005), sensitivity (P = 0.02) and specificity (P = 0.036). Application of fuzzy reasoning in an expert system for ultrasonography improves the diagnostic performance of inexperienced clinicians.