Detecting small blood vessels in color flow imaging: a statistical approach

Small blood vessels can be hard to detect with ultrasound imaging, because the blood-to-tissue signal ratio is low and the tissue component is strong. In this paper a model for the Doppler signal and a likelihood test for blood detection are presented. A maximum likelihood estimator for the blood Doppler frequency is presented and shown to be unbiased, while the commonly used autocorrelation estimate is biased for small velocities. A Neyman-Pearson probability test can be performed if the clutter spectrum and the blood Doppler frequency are known. Using a measured tissue signal and a modeled blood signal with narrow bandwidth, a theoretical minimum detectable blood velocity was found. Using a signal sampled from 8 pulses (a packet size of 8), a PRF of 250 Hz, a blood-to-tissue ratio of -40 dB and a tissue-to-noise ratio of 60 dB, the low velocity limit was found to be 4.3 mm/s for 6 MHz ultrasound. With a smaller packet size, the low velocity limit was higher, and for a packet size less than 4 the blood signal was not detectable