Detecting turbulent flow in Korotkoff sounds using nonlinear methods

Sounds generated by turbulent blood flow through partially occluded brachial artery were used to model sounds expected from coronary occlusions. Two methods of nonlinear analysis, third-order autocovariance and time reversal, were evaluated for their ability to detect blood flow turbulence. The values obtained from turbulent acoustic data were compared against surrogate data having similar power spectra but reduced complexity. Results showed that the values obtained by third-order autocovariance were significantly different from the surrogate data in all three subjects, but only when the artery was partially occluded and turbulent blood flow was present. The time reversal method also showed significant differences from the surrogate data in two of the three subjects. To evaluate the robustness of these approaches, noise equivalent to the standard deviation of the data was added. The third-order autocovariance method continued to show significant differences compared to surrogate data with little degradation in performance. These results indicate that third-order autocovariance provides a sensitive measure of the nonlinearities associated with turbulent blood flow though partially occluded arteries and may be of considerable diagnostic value.