Automatic detection of micro‐emboli by means of a generalized autoregressive conditional heteroskedasticity model
Résumé
Detection of micro‐emboli is of great clinical importance to prevent cerebro‐vascular events and to identify the causes of such events. Standard detection techniques implemented in the most commonly used systems are generally based on the passing of an energy threshold. The value of this threshold can be set just above the statistically highest detected energy of a Doppler signal related to circulating blood, i.e., during the systolic phase. This choice of threshold consequently prevents all detection of micro‐emboli events whose energy might be lower than the systolic energy. The main idea of our technique was to test adaptively the whiteness of the prediction error between the Doppler signal and a synthetic signal obtained by a GARCH model. By assuming that micro‐emboli are unforecasted, we hypothesize that embolic signals are unpredictable. We hypothesize that our new micro‐emboli detector is a candidate technique to detect very small micro‐emboli. We have tested and compared our new technique to the standard Fourier technique (Fourier). For a probability of detection fixed to 100%, the probability of false alarm of the standard automatic technique is 33% whereas the detector based on the GARCH model is less than 3%. This study demonstrates that our new technique detects micro‐emboli hitherto not identified by classical methods.
Origine : Fichiers produits par l'(les) auteur(s)