Automatic detection of microemboli by means of a synchronous linear prediction technique

Detection of microemboli is of great clinical importance to prevent cerebro-vascular events and to identify the causes of such events. As standard detection techniques implemented in the most commonly used systems cannot detect all of microemboli events whose energy is lower than the systolic energy, new techniques are proposed. By assuming that the Doppler signal is cyclostationary, we hypothesize that energy is statistically periodic. Furthermore, we hypothesis that embolic signals are unpredictable. Hence, the joint use of synchronous and linear prediction techniques could detect very small microemboli. If we periodically take and compare the values of the energy of the prediction error (or autoregressive parameters) at different time points in the cardiac cycle, we can therefore detect the presence of non-periodic events such as microemboli. In our study, we tested and compared our new technique to the standard technique (Fourier) using simulated and in vivo signals from patients with stenosis of high degrees of severity. From simulations, the standard automatic technique detected 60% of microemboli detected by our gold standard technique (audible detection and sonogram visualization) whereas the synchronous linear prediction technique detects 97% (the false alarm rate being set at 0%). From clinic examinations, the standard automatic technique only detects 67% of microemboli detected by our gold standard technique whereas the synchronous linear prediction technique detected 100%. This study demonstrates that our new technique detects microemboli up to now not identified by classical methods. Large microemboli are all detected, but small microemboli are only detected with our new technique. This latter technique opens up new prospects to detect small emboli, despite the need for further studies to incorporate "on line" technique.