Ovarian cancer diagnosis using a hybrid intelligent system with simple yet convincing rules

Abstract

Ovarian cancer is the ninth most common cancer among women and ranks fifth in cancer deaths. Statistics show that the five-year survival rate is greater than 75% if diagnosis occurs before the cancer cells have spread to other organs (stage I), but it drops to 20% when the cancer cells have spread to upper abdomen (stage III). Therefore, it is crucial to detect ovarian cancer as early as possible and to correctly identify the stage of the cancer to prevent any further delay of appropriate treatments. In this paper, we propose a novel self-organizing neural fuzzy inference system that functions as a reliable decision support system for ovarian cancer diagnoses. The system only requires a limited number of control parameters and constraints to derive simple yet convincing inference rules without human intervention and expert guidance. Because feature selection and attribute reduction are performed during training, the inference rules possess a great level of interpretability. Experiments are conducted on both established medical data sets and real-world cases collected from hospital. The experimental results of our proposed model in ovarian cancer diagnoses are encouraging because it achieves the most number of correct diagnoses when benchmarked against other computational intelligence based models. More importantly, its automatically derived rules are consistent with expert knowledge.