Improving the probability of clinical diagnosis of coronary-artery disease using extended kalman filters with radial basis function network

Abstract

Kalman filters have been popular in applications to predict time-series data analysis and prediction. This paper uses a form of Extended Kalman Filter to predict the occurrence of CAD (Coronary Artery Disease) using patients data based on different relevant parameters. The work takes a novel approach by using different neural networks training algorithms Quasi-Newton and SCG with combination of activation functions to predict the existence/non-existence of CAD in a patient based on patient’s data set. The prediction probability of this combination is resulted in accuracy of about 92% or above, using cross validation and thresholding to remove the limitation of time-series prediction introduced because of the Extended Kalman filter behavior.