Classification of pathological and non-pathological Cardiodynamicsgram (CDG) using nonlinear dynamics indexes

Cardiodynamicsgram (CDG) has emerged recently as a new noninvasive recording method of cardiac dynamics information during heart-beating process, which can be used for the early detection of myocardial ischemia. Automatic classification between pathological and non-pathological CDG can assist doctors in identification of the latent myocardial ischemia for their planning of minimizing ischemic time. In this work, we propose a classification methodology to classify pathological and non-pathological CDG. As the cardiac characteristics extracted by conventional morphology analysis methods is relatively limited and not comprehensive enough to reflect the dynamical characteristics of CDG, we extract three nonlinear dynamics indexes which include C₀ complexity (C₀), Approximate entropy (ApEn) and Lyapunov-exponent spectrum (LypE) as a measure of the complexity and nonlinearity of CDG in this work. These indexes are inputted into five different classifiers, including Naive bayes classifier (NBC), Support vector mechines (SVM), K-Nearest neighbour classifier (KNN), Multilayer perceptron (MLP), and Decision tree classifier (DTC). Experimental results at Chinese National Center for Cardiovascular Diseases show that encouraging classification accuracy can be achieved. Additionally, the computation time of these indexes extraction and classfication is very short, which is helpful for building up a real-time software tool towards assisting the physician in cardiology departments.