A Novel Nonlinear Predictability based Entropy Method for Evaluation of Noisy Signals

Benjamin Griffel, Mohammad K. Zia, Vladimir Fridman, Cesare Saponieri, and John L. Semmlow


Biomedical Signal Processing, Medical Signal Processing, Entropy, Nonlinear Analysis


Nonlinear biological signals such as heart rate and electromyogram data may appear noisy, but can contain information that is not apparent through linear analysis. Here we introduce a new measure, path length entropy (PLE), an entropy measure based on signal predictability. This entropy measurement characterizes the tendency of a signal to change trajectory. To evaluate this measurement, we compared PLE with Detrended fluctuation analysis (DFA), Multiscale entropy (MSE), Sample entropy (SampEn) and the Hurst exponent using a set of test noise samples and on heart sounds acquired at the chest. These samples included evenly distributed, Gaussian distributed and fractal noise of varying Hurst exponent. PLE was implemented both as a single analysis and also after a series of coarsening steps to probe for scaling effects. Analysis showed DFA to be most sensitive to each noise type, but PLE had nearly identical results and was 100 times faster. When applied to recorded heart sounds, PLE with scaling differentiated between normal and diseased subjects better than any other tested method. These results indicate that PLE may be a useful new measure of nonlinear biological signals and that multiple analysis methods may be needed to fully characterize nonlinear data.

Important Links:

Go Back