Time-frequency blood flow parameters estimation for micro-emboli detection

Detecting micro-emboli in the middle cerebral artery is a major challenge because these micro-emboli are related to the risk of stroke (cerebrovascular accident). Because micro-emboli are defined as High Intensity Transient Signals (HITS), their detection require the estimation of the blood flow background signal; i.e, the reference from which HITS have to be detected. In this paper, we propose an original micro-emboli detection procedure composed of two steps. Firstly, the background noise is isolated from the blood flow time-frequency image thanks to the log-likelihood segmentation method. Secondly, a HITS detection threshold is calculated from a blood flow statistic using the spectral kurtosis approach. Our procedure is used on the transcranial Doppler data from two patients and the results show that the mean percentage error of the velocity envelope estimation is reduced from 5% to 3% and the micro-emboli detection probability is 84% while only 49 artifacts are detected.