Blood detection performance in moving tissue

A method for evaluating the blood detection performance of general linear clutter filters is described. The detector performance is characterized by a receiver operating characteristic (ROC) which is a plot of the probability of detection, P/sub D/, versus the probability of false alarm, P/sub F/. With a Gaussian signal model, the optimal detector compares the power at the output of a clutter filter to a threshold. The optimal detector structure is thus similar to standard color flow systems, but with a filter matrix that depends on both the clutter and blood signal statistics. It is not possible to implement such a detector, but it gives the performance limit for practical detectors. The performance of a practically realizable adaptive clutter filter is evaluated. This filter compensates for the tissue movement by estimating the correlation matrix of the clutter signal by spatial averaging, and uses the eigenvectors corresponding to the largest eigenvalues as a basis for the clutter space in a regression filter. This basis gives maximum clutter attenuation for a given filter order. Digital RF data from the carotid artery was recorded, and a theoretical model for the blood signal was used to compare the detectors. With large tissue movement, the adaptive filter had almost optimum performance, and was significantly better than the polynomial regression filter.