Cardiac decision making using higher order spectra

Computer Aided Diagnosis for Cardiovascular Diseases Based on Ecg Signals: A Survey

“…For example, the hear rate variability is such a quantity of interest [25] • First and second order statistics -Mean and variance [26] • Discrete Wavelet Transform (DWT) -This feature extraction technique is closely related to spectrum techniques. Spectrum techniques provide only frequency restitution, whereas DWT provides both time and frequency resolution [27,28,27,29] • Independent Component Analysis (ICA) -The technique separates multivariate signals into their additive subcomponents [27,15,28] • Principal Component Analysis (PCA) -The statistical procedure is based on orthogonal transformation which produces linearly uncorrelated parameters known as the principal components [30,29,28,31,32] • Linear Discriminant Analysis (LDA) -Yields features that characterizes two or more signal classes [28] • Discrete Cosine Transform (DCT) -Spectrum technique based on cosine waves [15,33] Nonlinear methods are based on the more recent concepts of chaos and fuzzy logic [34,35,36]. The novelty of these methods is reflected in the fact that only a few recent CAD systems employ them, as indicated in the following list.…”

“…Used to describe the fractal nature of a signal [37,38] • EM Clustering -The expectation-maximization (EM) algorithm constitutes an iterative method that estimates parameters in statistical models [39] • Higher Order Statistics -Statistical measures beyond mean and variance [31,32] …”

“…the signal classes a linearly separable. For nonlinear data, nonlinear kernel functions are used [49,28,29,31,32].…”

“…• K-Nearest Neighbour (K-NN) -The majority of neighboring known signal points decides the signal class of an unknown input [15,28,31].…”

“…• Classification And Regression Tree (CART) -Used to construct prediction models form ECG data [53,31] • Random Forest (RF) -Ensemble learning algorithm [31] • Threshold -Binary decision based on a threshold value [25,26,22,23].…”