In medical applications ultrasound has been successfully used for diagnostic purposes of the human circulatory system for more than 40 years. Color-Doppler ultrasound systems are standard equipment of every hospital, because of their safe use and the immediately available results.

This work describes improvements of signal processing algorithms used in Color-Doppler ultrasound systems to map blood flow spatially and temporally resolved in three dimensions. In small vessels it is a very demanding task to estimate the flow velocity because the scatterers are very slow moving and show rather small scattering cross sections. In addition motions issuing from heart action, respiration and tremor are superimposed to the flow motion. Only the relative velocity between blood flow and vessel wall is of interest, because they are equivalent to the real blood flow velocity. A precise velocity estimation is difficult, if the magnitude of the wall velocity is much higher than the flow velocity, since in this case the standard deviation of the wall velocity estimator reaches a magnitude comparable to the blood flow velocity to be estimated.

All necessary signal processing steps are discussed such as signal alignment, clutter filtering, velocity estimation, noise rejection, and three dimensional reconstruction of the flow field. The developed methods have been applied to real flow data, acquired by a developed ultrasound system, that operates with shorter acoustic wave lengths than commercial devices. In the first step the measurement accuracy has been experimentally verified with a special flow phantom, which shows perfectly tuned geometrical and dynamical properties. Additionally, in vivo measurements have been performed to proof its serviceability.

This work shows visualizations of velocity fields in vessels of an inner diameter of less than 0.5 mm. A velocity resolution < 1 mm/s will be achieved, if a wave length of < 0.1 mm is utilized for the interrogation.

Introduction

Ultrasonic scanning systems due to their non-invasive and non radiative properties are very wide spread in medical imaging. Interrogation of tissue allows to detect deviations from normal vascularization indicative of malignant processes for example in breast tissue [1]. This is done by detecting small to very small blood vessels and determining their 3D structure which in case of malignancy would show arterial loops or arterio-venuous shunts for example [1], [2]. Furthermore the onset of vascular disease can be detected if flow profiles with high spatial, temporal and flow velocity resolution can be measured.

Experimental Set-up

In order to be able to verify the accuracy and variance of the velocity estimation procedure described later on a phantom was designed and built that is able to simulate both pulsatile flow and vessel motion that is known to occur in in-vivo experiments. An infusion pump is used to provide volumetric flow in the range of 6 mm³/min up to 6 cm³/min through a simulated blood vessel in the phantom. The measurement set-up is built from standard components like a digital oscilloscope, a pulser/receiver some signal generators to provide synchronization capabilities and a transducer (center frequency f0 = 15 MHz).

This schematics not only shows all hardware components of the set-up but also indicates how the devices are interfaced. The hardware consists of the following parts: a standard PC currently running Matlab to do all the signal and image processing in the development phase (where processing speed isn't the main goal) besides controlling the translation stage via an RS-232 link. A digital scope (TEKTRONIX TDS 744A) used as transient recorder and two signal generators (HP 33120) are connected to the PC via a fast USB-GPIB bus interface. One signal generator controls the pneumatic excitation system and the other simulates an adjustable delay line. Furthermore a transducer of type PANAMETRICS V319, that has a spherical acoustic lens with a focal length of 25.4 mm and a center frequency f0 of 15 MHz is mounted onto the 3D-translation stage, that is capable of moving the transducer in x-, y- and z-dimensions in increments of 10 mm. An ultrasonic analyzer model PANAMETRICS 5052UA capable of pulse repetition rates between 100 Hz and 5 kHz, and pulse peak voltages of up to 300 V with rise times less than 10 ns (10 % to 90 %; no external load) is used [4].

The flow phantom (shown in Fig. 1) consits of a single flow channel with an inner diameter of 0.5 mm.

Signal and Image Processing

The signal processing necessary to obtain reliable velocity estimates even under severe inherent motion is rather involved. It is basically based on a modified phase shift estimator originally proposed by Jensen [3]. In Fig. 2 the sequence of signal processing steps discussed in this section is shown. At first overall tissue motion with respect to the transducer is determined for each ultrasonic pulse followed by a digital realignment to keep presumed stationary targets (like vessel walls) at constant depth. This is an absolutely necessary processing step, since flow velocities in small arteries can be as slow as 0.1 mm/s whereas the subjects motions can be as fast as 1.35 mm/s [2].
The present results of the reconstruction of a 3D velocity field as seen for example in Fig. 3(e) obtained by isonifying our flow phantom under non-stationary flow conditions. To compile the flow field 45 slice planes with 20 lines of sight each were interrogated, their flow estimated and the result put into a 3D data structure suitable for MATLAB's volume rendering capability. Fig. 4 shows the volumetric velocity data at one instant in time approx. in the middle of the simulated heart cycle i=43 (86 ms after acquisition start). The slices cut through the isosurface show the velocity field gray scaled. Note the changes of the velocity magnitude along the y-axis between 5 mm to 8 mm and the artefacts stemming from fast moving tissues in the depth range z of -28 mm to -29 mm. Our estimation algorithm wasn't able to remove these artefacts, because the tissue velocity characteristic was similar to flowing blood.
Fig. 3(a) shows the raw data as a brightness mode image. The signal processing steps such as alignment, wall-filtering, velocity estimation and noise rejection are independently performed for each LOS.