A semiautomatic method for obtaining a predictive deep learning model and a rule-based system for abdominal aortic aneurysms

Development in the medical field is getting fast every day. People's interest in improving their expectancy of life, their life quality, and the significant investments in medical laboratories modify the diagnosis methods, application protocols, and surgical techniques. One of the most significant milestones in the medical field have been incorporating computers to improve data analysis during the last years. Nowadays, it is a fact that computers can help physicians, i.e., the use of artificial intelligence techniques. This paper proposes a multistage prediction-based approach and a rule-based system for the treatment of abdominal aortic aneurysms. The first step is to develop a neural network model to predict 30-day mortality during and after aortic endovascular procedures. The second step aims to infer a rule-based system from the previous model. The results show that with only eight features the final predictive model can obtain an accuracy of around 87%. Furthermore, a decision tree with the same accuracy can be inferred from this model using three features and four rules