Fuzzy Logic for Medical Diagnosis

Abstract

Medical diagnosis is a complex process due to the complexities, uncertainties and vagueness of the symptoms involved, and sometimes because of their indirect relationship with the final output. Traditional systems for diagnosis very often incorporate certain inabilities that eventually lead to the vagueness in the diagnosis result. Besides this, imprecise and incomplete knowledge are difficult for these traditional disease diagnosis expert systems to analyze. The fuzzy logic has carved a niche in medical diagnosis, for its ability to handle the dynamic nature of the disease diagnosis and medication. Various approaches of Fuzzy Logic, namely, Type-1 Fuzzy Logic, Interval Type-2 Fuzzy Logic, and General Type-2 Fuzzy Logic are being used for decision making in medical diagnosis. The fuzzy rule base is what that makes these approaches stronger. Various expert systems using these methodologies have been designed, however, an extensive study about how these different fuzzy based approaches serve the medical diagnosis that very often an evolvement of the basic Type-1 FL is demanded, leading to the formulation of ‘layered Type-1 FL’ approach called Interval Type-2 Fuzzy Logic, needs to be conducted to mathematically analyze the difference in the functioning of these approaches. In the thesis, these two approaches, i.e., Type-1 Fuzzy Logic and Interval Type-2 Fuzzy Logic, are implemented on disease dataset and are compared in terms of the accuracies of their predictions for two prominent lifestyle diseases, namely, diabetes and heart related complications. Type-1 Fuzzy Logic performs fuzzification using trapezoidal membership function, then rule inference and rule aggregation using MAX fuzzy based disjunction method is performed, and then the defuzzification using the centroid method. Interval Type-2 Fuzzy Logic performs type-2 fuzzification using the trapezoidal membership function with a uniform FOU, then rule inference is performed using MAX fuzzy based disjunction method. After that type-reduction using Karnik-Mendel algorithm computes the type-1 rule aggregation output, and finally the defuzzification using the centroid method computes the final output, i.e. probability. In this way, a disease specific performance of these fuzzy based methodologies is studied which is helpful to understand their usage and the functioning distinctively