The filter used to separate blood signals from the tissue clutter signal is
an important part of a color flow system. In this paper, statistical detec
tion theory is used to evaluate the quality of the most commonly used clutt
er filters. The probability of falsely classifying a sample volume as conta
ining blood is kept below a specified threshold. With this constraint, the
probability of correctly detecting blood is calculated for all the filters.
Using a measured clutter signal, we found that polynomial regression filte
rs and projection-initialized IIR filters are best among the commonly used
filters. The probability of correctly detecting blood with velocity 10.1 cm
/s was 0.32 for both these filters. The corresponding value for the optimal
detector was 0.81, whereas a regression filter that depends on the clutter
signal statistics achieved a blood detection probability of 0.72. (C) 2000
Elsevier Science B.V. All rights reserved.