Adaptive methodology for designing a predictive model of cardiac arrhythmia symptoms based on cubic neural unit

Abstract

A cubic neural unit is a kind of a higher-order neural unit which can be used for prediction tasks among others, in the medical field. The example of the tasks includes monitoring cardiac behavior in real-time either for preemptive treatment, or for supporting a doctor to reach a more accurate diagnosis. We propose a predictive model which has been developed as an application in open source code with the aim to make it publicly accessible for research community and medical professionals and also to decrease the implementation cost. The proposed model uses sample-by-sample adaptation of the gradient descent method with error backpropagation. This paper presents an application of a cubic neural unit as a prediction mechanism for abnormal cardiac behavior, and it describes a new adaptive methodology based on application of a dynamic cubic neural unit for cardiac arrhythmia prediction. To validate the model, it has been tested on the data from the Massachusetts Institute of Technology-Beth Israel Hospital Cardiac Record Database. This paper is focused on premature ventricular contraction, atrial premature contraction and normal heartbeat records.