USE OF CLASSIFICATION MODEL WITH ONTOLOGY TO DEVELOP MEDICAL KNOWLEDGE BASED SYSTEM FOR BREAST CANCER DIAGNOSIS AND TREATMENT RECOMMENDATION

Abstract

Among the various types of cancer diseases, breast cancer is the first-ranked cause of death in Ethiopia. Considering this, the researcher attempts to use a classification model with ontology to develop a medical knowledge-based system for breast cancer diagnosis and treatment recommendation. To conduct this study, the researcher used an ontology-based design science research methodology. Ontology was developed to share and re-use the knowledge of domain experts. The use of ontologies in the medical domain has gained wider acceptance in recent years and has been accompanied by great success, which motivated this study to create ontologies for the diagnosis of breast cancer that serve as a knowledge base. In this study, the researcher used classification algorithms such as the J48 decision tree, decision tables, and JRIP to extract rules from the breast cancer dataset and selected J48, which performed well, and transformed the extracted rules to SWRL that infer new knowledge from domain ontology for representing the knowledge needed to diagnose breast cancer to its stages. Ontology uses DT rules, SWRL rules, and the Drool inference engine to classify breast cancer stages. The Protégé tool that supports OWL representation is used to build concepts and relationships in an ontology for knowledge representation. The constructed ontology is used for knowledge structuring and representation in the form of hierarchy, and a top-down approach is used to express domain knowledge. Similarly, the medical knowledge-based system is developed based on the rules extracted from the breast cancer dataset using a classification algorithm and later mapped to SWRLs to construct an ontology model. The proposed prototype analyzes the symptoms of the patients and gives the exact stages of breast cancer and the appropriate recommended treatments. Finally, the system performance testing is done by using test queries, which achieved an accuracy of 98.67%, precision of 96%, recall of 100%, specificity of 98.94% and F-measure score of 98%. Likewise, user acceptance testing also takes place and achieves 93.8% user acceptance. This indicates that the developed prototype is promising to come up with an applicable system. To enhance the performance of the prototype, there is a need to use an image-based breast cancer dataset and apply deep learning algorithms, machine learning, and hybrid techniques to increase the accuracy of the model.