Smooth bandpass empirical mode decomposition with rolling ball sifting for extracting carotid bruits and heart sounds

Carotid bruits are systolic sounds associated with turbulent blood flow through atherosclerotic stenosis in the neck. They are audible intermittent high-frequency sounds mixed with low-frequency heart sounds that wax and wane periodically. It is a nontrivial problem to extract both bruits and heart sounds with high fidelity for further computer-aided analysis. In this paper we propose a smooth bandpass empirical mode decomposition (EMD) method to tackle the problem in the time domain. First, bandpass EMD is achieved by using a rolling ball algorithm to sift the local extrema of chosen time-scales. Second, the local zero is smoothed by interpolation with a monotone piecewise cubic spline. Preliminary results indicate that the new method is able to extract both carotid bruits and heart sounds as visually smooth oscillating components.