Prediction of Coronary Artery Disease Using a Combination of Methods for Training Radial Basis Function Networks

Cardiovascular disease (CAD) is among the most prevalent diseases around the world; nevertheless, its diagnosis requires highly qualified medical staff (e.g., cardiologists) because of the many variables involved in the process. Due to diagnostic complexity and the limited number of available qualified staff, the development of smart systems that could automate the diagnostic process is paramount. This paper investigates two systems in order to achieve this goal. The first system proposes the application of a data fusion with Kalman filtering in diagnosing CAD as well as in the prediction of the need to conduct a Coronary Artery Bypass Graft (CABG) in patients identified as having CAD.The second system, which is based on a combination of Particle Swarm Optimization (PSO) and a Gravitational Search Algorithm (GSA), is also proposed. Patient data was gathered from King Abdullah Medical City in Saudi Arabia, and a statistical analysis was conducted to explore the relationship between an array of variables and CAD. After identifying pertinent variables for diagnosis, some learning algorithms (e.g., Kalman Filtering, Particle Swarm Optimization and Gravitational Search Algorithm) were applied to the collected data sets to train the system for predicting the diseased condition. The main aim of this paper is to identify the underlying functional relationship between the medical patient records and the medical diagnosis in the datasets in order to predict the presence or absence of the disease for new patients. This work takes a novel approach by using different neural networks training algorithms, e.g., Quasi Newton and Scaled Conjugate Gradient (SCG) with several activation functions on an extended Kalman filter.