Micro-emboli detection: an ultrasound Doppler signal processing viewpoint.

Several studies have been carried out in the last twenty years on the characterization and detection of cerebral artery emboli. From the detection point of view, the existing methods are largely based on the classical Fourier analysis of which the well known limitations provide poor accuracy. This paper first recalls existing methods based on Fourier, Wigner-Ville and wavelet approaches. It then presents new emboli detection methods based on parametric signal processing approaches. The basic idea of these parametric methods is to compare the Doppler embolic signal to its autoregressive model. The detection principle consists in constructing a decision information which contains the signature of the micro-embolus being sought. The detection is finally evaluated using receiver operating characteristic (ROC) curves. Comparison between the new methods and classical approaches is performed using a realistic embolic signal simulation. Furthermore, to validate our theoretical study, we tested our new algorithms using in vivo signals. This comparison shows the significant inaccuracy of existing methods to detect micro-emboli.