A number of clinical applications require estimating blood velocity. Fig. 1 illustrates a flow motion within a vessel.
Fig. 1
Ultrasonic imaging is often used to reach this goal in real-time imaging. Doppler techniques are a reference to estimate blood velocities. However, Doppler suffers from a number of limitations:
- poor estimation of low velocities
- the spatial resolution is limited Moreover, only the axial component of the velocity is estimated.
Consequently,
- the flow orientation must be known to estimate the velocity modulus
- and the angle between the flow and the probe into the imaging plane must be smaller than 60°.
In order to overcome these limitations, we develop alternative methods to estimate blood flow velocity. Two ways are currently the objects of research. The first one deals with a statistical approach of the moving sequence of ultrasound images. The second one deals with a spatiotemporal approach and considers the sequence of US data as a 3D volume.
Statistical approach
Spatiotemporal approach
A temporal sequence of 2D images can be seen as a 2D+t spatiotemporal volume. A sequence in translation leaves a trace in this volume as we can see it in Fig. 2. The vessel represented into the imaging plane $(x,y)$ is delimited by the white edges and consisted of moving scatterers. The plane $(x,t)$ contains an oriented texture and its orientation is related to the velocity of the scatterers as in Eq. (1) :
$\displaystyle{v=\frac{F_t}{F_s}tan\,\theta}$ (1)
Velocity is local and consequently we have to estimate orientation in each pixel of the sequence. In
Fig. 2 | Fig. 3 |
Our method was validated on a large set of sequences
Fig. 4 | Fig. 5 |
Let us add that the bank of filters presented in Fig. 3 is dedicated to estimation of longitudinal flows. The use of 3D spatiotemporal filters is necessary to estimate vector-velocity of non-longitudinal flows.