Comparison among fractals features to differentiate classes of ages using segments of the ECG signal

Abstract

This report describes the application of 9 fractal-dimension features to 3 12-lead ECG signal segments. The resulting values ​​were used to train and evaluate MLP classifiers to: 1) determine the extent to which the fractals capture age-related changes in the ECG signal; and 2) determine which of the three signal segments might contain the chronological information extracted by the fractals. Four age-derived classes are used, with varying degrees of specificity.Fractals of Higuchi, Katz, line length, Normalized Length Density, Petrosian, Sevcik, Power Spectral Density slope, Correlation Dimension, Hurst were used; applied to each of the 12 leads. The segments were defined taking the fiducial point R as a reference: segment PR; segment RT; and segment PQRST. As a result, accuracy rates of 75% were obtained over two age classes (threshold of 50 years) from the development of an MLP classifier trained with the average values ​​and medians of the 9 combined features. No differences are recorded when using one or another signal segment, among the three defined.