Reliability evaluation of emboli detection using a statistical approach

Cerebral emboli detection is a problem of considerable practical importance for monitoring vascular pathologies. If in most practical in vivo or in vitro experiments, big-size emboli detection seems comfortable, small-size emboli detection remains a challenge. The aim of this paper is to evaluate the performances of micro-emboli detections using statistical tools. The authors first introduce for this purpose a specific decision information through Autoregressive (AR) modeling. To overcome the problem of empirical detection threshold choice and quantify the reliability of the emboli detection, the chosen decision information makes it possible to link the threshold with the probability of false alarm (PFA). The authors then extend this approach to classical detection methods based on nonparametric analysis in order to perform comparisons, by using computer simulation of embolic signal. It is shown that reliable detection is not possible by the classical approaches when the Embolus-to-Blood Ratio (EBR) is lower than 10 dB since in this case the PFA is by far above 10%, whereas the authors' approach can detect small-size emboli with EBR as low as 5 dB since the PFA is then below 5%. In vivo measurements are finally performed in order to validate the authors' approach.