Ontology enhanced fuzzy clinical decision support system

Abstract

Abstract A hybrid ontology-based fuzzy decision support system (OBFDSS) has been presented to support physicians in diabetes diagnosis problems. This work combines fuzzy logic and ontology to have both capabilities of representing semantic knowledge and reasoning with vagueness. Such a CDSS encodes the medical knowledge of the diabetes experts’ backgrounds or common reasoning knowledge regarding ontology. The ontology concepts, relationships, properties, and axioms are based on literature studies and domain expert experience. Ontology is used to represent the semantic structure as well as to calculate the clinical similarity between the compared diseases, medications, and symptoms related to diabetes. Then, the level of similarity is passed to a fuzzy inference engine. Our contribution is the development of a semantically intelligent fuzzy expert system. We will enhance the inference capabilities of the regular fuzzy system with the OWL2 ontology reasoning capabilities. The methodology of the framework can be defined in four stages: knowledge acquisition and features definition, semantic modeling, fuzzy modeling, and knowledge reasoning. The knowledge acquisition and features definition stage is responsible for acquiring the knowledge for diabetes diagnosis. The diabetes knowledge is collected from medical experts, the recent literature, EHR databases, and recent CPGs. Semantic modeling is related to the ability to represent semantic features in the form of ontology and to measure the level of clinical similarity between compared medical concepts. We propose an OWL2 ontology based on SNOMED CT standard medical terminology. The fuzzy model includes the definition of the fuzzy features and its associated fuzzy terms and fuzzy rules. Three types of knowledge including CPGs, the experience of domain physicians, and the processing of EHR data are defined in the fuzzy model. There are several issues that must be resolved in knowledge reasoning to generate fuzzy rules, enhance the generated fuzzy knowledge, utilize the fuzzy inference engine, and complete defuzzification. The proposed hybrid model offers a powerful tool in the diagnosis of diabetes. It involves building a complete linguistic fuzzy rule base that can integrate knowledge from experts and CPGs with knowledge extracted from training data as well as from the semantic model. This step enhances the level of automation and interoperability of CDSS. We expect to achieve acceptable and more flexible performance when using standard medical terminology. In this work, a hybrid OBFDSS is presented. The hybrid model is the most logical step to improve the fuzzy expert system by adding a semantic reasoning process to its capabilities. We expect that our proposed hybrid system will emphasize the significance of combining ontology and Mamdani fuzzy inference for diagnosing diabetes.