Entropy-Based Nonlinear Analysis for Cerebral Emboli Detection Using Doppler Signals

Emboli in blood vessels may be of different sources, and recordings from used instruments define the distribution of various components. Flow examination with Transcranial Doppler ultrasound in the cerebral hemispheres provides a way to detect emboli during medical procedures. Different features represent the nature of components. Complexity in uncorrelated and correlated channels is spanned with probabilities of variables. With growing interest in the application of nonlinear methods to determine the uncertainty in time series signals, entropy-based methods are better suited to understand the dynamics of time series data. In this paper, we analyzed a set of Doppler signals collected from patients reported with carotid stenosis and characterized the detection and classification of Doppler emboli signals, artifacts, and Doppler speckle. We performed Doppler signal characteristic understanding using entropy measures that include approximate entropy, sample entropy, permutation entropy, and dispersion entropy. With introduction related to different entropy measures, we present different numerical and graphical cerebral emboli detection results with more focus on dispersion entropy as it works better for long Doppler signals. Results presented in this work suggest that emboli signals are well differentiated from artifact and Doppler speckle signals using dispersion entropy at embedding dimension value of m= 2 (mean±std.=3.05±0.65) and dispersion entropy class value of c= 3 (mean±std.=5.25±2.54).