1 3
Exp Fluids (2017) 58:3
DOI 10.1007/s00348-016-2283-9
REVIEW ARTICLE
Ultrasound Imaging Velocimetry: a review
Christian Poelma
1
Received: 22 August 2016 / Revised: 6 November 2016 / Accepted: 7 November 2016 / Published online: 15 December 2016
© The Author(s) 2016. This article is published with open access at Springerlink.com
being laser-Doppler anemometry (LDA) and particle image
velocimetry (PIV). While very successful for single-phase
flows in e.g. wind tunnels or water flumes, the presence of a
dispersed phase with a volume fraction as low as 0.5% can
render these techniques useless (Deen et al.
2002; Poelma
et al. 2006). The exact limiting volume fraction depends on
the dispersed phase size, distribution and the dimensions
of the flow domain (Linne et al. 2009). For small particles,
droplets, and bubbles the limit is reached earlier, while it
may be possible to obtain results ‘in between’ large bubbles
with an overall large volume fraction (Mudde et al. 1997).
The latter approach case may lead to biased statistics, how-
ever. For some studies, it is feasible to use refractive index
matching, so that the system remains optically transpar-
ent (Wiederseiner et al. 2011). Unfortunately, only a lim-
ited number of combinations of solid and liquid materials
can be used to achieve this matching, severely limiting the
physical parameter space that can be probed. This means
that many important two-phase flow problems, inspired by
e.g. industrial or geophysical applications, are out of reach
for the current optical measurement techniques.
In the last decade, a series of non-optical imaging
modalities have been introduced for flow measurement,
most of them based on medical imaging techniques. Exam-
ples include Magnetic Resonance Imaging (Elkins and
Alley 2007; Ooij et al. 2011; Lakshmanan et al. 2016),
X-ray imaging (Fouras et al. 2007; Heindel 2011) and
ultrasound imaging (echography). Alternative strategies use
tomographic reconstruction of electrical signals from sen-
sors surrounding the flow, such as electrical capacitance
tomography and electrical impedance tomography (ECT,
EIT) (Yang and Peng 2002; Dyakowski et al. 2000; George
et al. 2000). This review focusses on a particular subset of
ultrasound techniques, namely correlation-based velocity
estimation using echography image data; this is distinct
Abstract Whole-field velocity measurement techniques
based on ultrasound imaging (a.k.a. ‘ultrasound imaging
velocimetry’ or ‘echo-PIV’) have received significant atten-
tion from the fluid mechanics community in the last dec-
ade, in particular because of their ability to obtain velocity
fields in flows that elude characterisation by conventional
optical methods. In this review, an overview is given of the
history, typical components and challenges of these tech-
niques. The basic principles of ultrasound image formation
are summarised, as well as various techniques to estimate
flow velocities; the emphasis is on correlation-based tech-
niques. Examples are given for a wide range of applica-
tions, including in vivo cardiovascular flow measurements,
the characterisation of sediment transport and the charac-
terisation of complex non-Newtonian fluids. To conclude,
future opportunities are identified. These encompass not
just optimisation of the accuracy and dynamic range, but
also extension to other application areas.
1 Introduction
The de facto standard non-invasive flow measurement tech-
niques in research laboratories are based on optical prin-
ciples (Tropea et al. 2007), the most prominent examples
Electronic supplementary material The online version of this
article (doi:
10.1007/s00348-016-2283-9) contains supplementary
material, which is available to authorized users.
* Christian Poelma
c.poelma@tudelft.nl
1
Laboratory for Aero and Hydrodynamics (3ME-P&E),
Delft University of Technology, Leeghwaterstraat 21,
2628 CA Delft, The Netherlands
Exp Fluids (2017) 58:3
1 3
3 Page 2 of 28
from Doppler-based methods, such as e.g. Colour Doppler
or Ultrasound Doppler Profiling. For clarity, these Dop-
pler methods are also shortly described. For an extensive
review of these techniques in a clinical setting, the reader
is referred to the recent two-part review by Jensen et al.
(
2016a, b). Correlation-based techniques are known under
various names in literature, including Ultrasound Imag-
ing Velocimetry (UIV), speckle tracking velocimetry, and
echo-PIV. They all refer to methods that apply PIV data
processing techniques to data that has been obtained using
echography (i.e. ultrasound imaging). Despite the alterna-
tive imaging modality, UIV retains the main strength of
PIV: it produces non-invasive, instantaneous, two-dimen-
sional, two-component velocity fields. An illustrative appli-
cation example is shown in Fig. 1.
Within the imaging modalities, ultrasound-based tech-
niques represent a good compromise between imaging/
measurement capabilities and practical considerations and
constraints. For instance, it is significantly more affordable
than MRI, which is also bound to a specific site (void of
ferromagnetic materials) and which requires more com-
plex postprocessing of data. Ultrasound-based techniques
do also not require the stringent radiation regulations sur-
rounding X-ray imaging. Furthermore, the latter is a shad-
owgraphy technique: the signal is integrated along the
beam axis, complicating the analysis of three-dimensional
flows. A tomographic approach is a possible solution
(Fouras et al. 2007), but this is suitable for time-averaged
flow patterns only. Optical coherence tomography (Huang
et al. 1991) has a superior resolution, but is restricted to a
small field-of-view. Finally, the resolution of ultrasound
surpasses that of ECT/EIT, which generally uses a limited
number of sensors. With this limited amount of informa-
tion, reconstruction of complex phase distributions within
the measurement domain becomes difficult, as the inverse
problem at the heart of the method is severely ill-posed.
The accessibility, relative simplicity and low cost, com-
bined with ever increasing possibilities, has resulted in a
strong increase in the use of UIV in the last decade. Appli-
cations range from in vivo cardiovascular flow studies to
more traditional fluid mechanics studies. In this manu-
script, the basic concepts of ultrasound imaging (Sect. 2)
and velocity measurement (Sect. 3) are reviewed. While
UIV borrows heavily from PIV, there are some fundamen-
tal differences; these are described in Sect. 4. Examples
of UIV applications are given in Sect. 5. Conclusions and
some future opportunities are given in Sect. 6.
2 Ultrasound imaging
As this review is primarily intended for the fluid mechan-
ics community, a brief historical overview (Sect. 2.1) and
a basic introduction into ultrasound imaging (Sect. 2.2)
will be given. Excellent in-depth discussions are found in
e.g. Szabo’s monograph Szabo (2004) and various medical
imaging textbooks, e.g. Suetens (2009).
2.1 Historical perspective
Ultrasound, defined as sound with a frequency above 20
kHz, has been used for diagnostics since the 1940s (Szabo
2004). While ultrasound was an established phenomenon
in the nineteenth century, practical ways to harness it were
enabled by the discovery of the piezo-electric effect by
Jacques and Pierre Curie in 1880–1881 (Curie and Curie
1881). Research built on work in the field of SONAR,
which emerged during the First and Second World Wars,
and was used to locate submerged objects using echo-
location. One of the earliest peacetime applications was
an apparatus to detect flaws in metals—the ‘Supersonic
Reflectoscope’ by Firestone (1946). The device made use
Fig. 1 An illustrative example of Ultrasound Imaging Velocimetry.
(Left) a linear transducer placed at the wall of a pipe containing an
opaque flow (here a strongly non-Newtonian drilling fluid). Repro-
duced from Poelma and Gurung (
2016). (Right) the resulting vector
field at the centerplane, superimposed on an instantaneous ultrasound
image Reproduced from Gurung et al. (
2016) ©IOP Publishing.
Reproduced with permission. All rights reserved
Exp Fluids (2017) 58:3
1 3
Page 3 of 28 3
of the property of ultrasound that it can penetrate non-trans-
parent materials: first, a short pulse of ultrasound waves
was created using a quartz-crystal and sent through a test
sample. If a flaw was present in the sample, the variation
in the material properties (in this case impedance) resulted
in a reflection of a part of the sound waves. The intensity
and time delay of these reflections, recorded using the
same crystal and visualised on an oscilloscope, could then
be used to localise and quantify the flaw. Around the same
time, Dusik pioneered medical diagnostics using ultra-
sound (Dussik
1942). In his earliest experiments, he used
a transmission approach (i.e. placing the receiving crystal
beyond the field of interest) to study the human brain; this
led to the first ever ultrasound image recorded.
At the core, modern echography still operates along
the same lines as Firestone’s device. Modern devices use
piezo-electric transducers to create and receive ultrasound.
The received signals are naturally no longer directly dis-
placed on an oscilloscope, but are digitised and processed
further to form an image. Initially, only a single transducer
was used. The 2D image created by Dussik was obtained
by manually translating the single transducer (i.e. a point-
by-point scan). Later, freehand scanning was introduced,
which tracked the position and orientation of the trans-
ducer; by combining data from various viewing angles,
this allowed reconstruction of stationary images in 2D
and, later, in 3D (Barry et al. 1997). Mechanical scanners
introduced faster imaging rates, so that (successive) 2D
snapshots could be obtained. This was achieved by encas-
ing a linear translation mechanism for the transducer ele-
ment (Wild and Reid 1952). Alternatively, the transducer
could be revolved or rocked around its axis, leading to sec-
tor imaging (with its characteristic wedge-shaped field of
view). Developments in transducer materials, design and
electronics led to phased-array probes (Bom et al. 1973;
Ramm and Thurstone 1976). These phased-array trans-
ducer form the basis of modern equipment. They con-
struct images by sequentially reading out elements, rather
than mechanically translating an element. This has addi-
tional advantages, such as the possibility of beam steer-
ing and focussing. As a linear phased-array creates a 2D
image, a 3D image can be constructed by sweeping the
probe (Fenster et al. 2001). Alternatively, matrix trans-
ducers have been introduced (Smith et al. 1991) to obtain
3D data. Apart from such hardware developments, a bet-
ter understanding of ultrasound wave physics and more
sophisticated signal processing techniques have led to
better image quality; a prime example is harmonic imag-
ing (Duck 2002). In this method, the original frequency
is filtered out and only non-linear responses are used for
image reconstruction. This way, particular structures can be
highlighted that would otherwise be obscured (e.g. contrast
microbubbles; see later). Furthermore, the introduction of
plane-wave imaging has increased the imaging rate sig-
nificantly (Tanter and Fink 2014). In this technique all
transducer elements are used simultaneously (rather than
sequentially) and the image formation is done off-line
using the recorded signals, analogous to digital holography
(Tanter and Fink 2014; Schnars and Jueptner 2005). Apart
from these technical breakthroughs, there have been many
ultrasound innovations that are application-driven, such as
the development of miniaturised transducers for intravascu-
lar ultrasound imaging (‘IVUS’, Roelandt et al. 1989) and
specific transducers for specific organs and/or procedures
(Szabo and Lewin 2013).
For this review, the focus is on conventional, linear
phased-array transducers, as these are common in experi-
mental fluid mechanics. In Sect. 6, the implications of
some new developments on UIV will be discussed. A sche-
matic representation of a linear phased-array transducer is
shown in Fig. 2. Typically, these transducers have 64 up to
512 elements, which can emit and receive independently in
specific patterns for a wide range of imaging modes. Each
transducer has a certain inherent centre frequency (based
on the resonance frequency of the particular piezo-electric
element design), but can operate in a wider range (band-
width), e.g. 5–10 MHz. Figure 2 also introduces the con-
ventional coordinate system and terminology, which will
be used and explained in subsequent sections.
2.2 Ultrasound imaging basics
Figure 3 demonstrates the basic principles for the acquisi-
tion of an ultrasound image. Consider one transducer ele-
ment of a phased-array. An alternating electrical signal is
applied to this element, typically made of PZT (lead zirco-
nate titanate) or PVDF (polyvinylidene fluoride) (Suetens
2009). The alternating signal leads to a deformation of the
piezoelectric element and thus acts as a source of a pressure
Fig. 2 Schematic representation of a linear phased-array transducer
and coordinate system. ABW azimuth beam width
Exp Fluids (2017) 58:3
1 3
3 Page 4 of 28
wave. Generally, relatively short pulses are created, with a
frequency f. For medical applications, generally frequencies
in the range of 1–10 MHz are used. High-frequency devices
exist for specific applications, which operate in the 30–100
MHz range (Lockwood et al.
1996). For simplicity, it is here
assumed that this pressure pulse is a small-amplitude, longi-
tudinal compression/rarefaction wave travelling along the z
(beam) axis, i.e. perpendicular to the transducer array.
The longitudinal wave travels with a velocity c, the
speed of sound of the medium that carries the wave. If
this wave encounters a medium with different properties,
it may partially be transmitted and partially be reflected.
The strength of the reflection is given by the difference in
the specific acoustic impedance, as discussed in detail in
the next section. In the schematic of Fig. 3a, the pulse first
hits the surface between gel and the wall containing the
flow. This creates a strong reflection, which is recorded
by the same transducer element. Figure 3b shows the
voltages (corresponding to pressure variations) from the
transducer element as a function of time. At around 15 ms
the strong echo from the wall can be observed. By using
the speed of sound, the time axis of the collected data
can be converted into a spatial axis, as shown in Fig. 3c.
This type of data is called A-mode, with ‘A’ referring to
amplitude. Several additional steps were performed going
from (b) to (c): a Hilbert transform is used to remove the
fast modulation on top of the signal (also referred to as
envelope detection), see also the small inset in subfigure
(b). Furthermore, the data is log-compressed, to reduce
the difference in intensity of specular reflections (e.g.
from flat surfaces) and scattered signals (e.g. blood, tracer
particles).
Once the data from this particular transducer element
has been collected, the process can be repeated using the
subsequent element. The result is a 2D ultrasound image,
as shown in Fig. 3d. The dashed vertical line represents
the data shown in subfigure (c); each vertical line is also
referred to as a scan line. These 2D results are referred to
as B-mode images, with b referring to ‘brightness’. Alter-
natively, the same transducer element can be read out again
and again. This creates M-mode data, where the vertical
axis represents the z position (distance from the transducer)
and the horizontal axis represents time. This mode is often
used to study fast dynamic effects in the body, such as e.g.
the motion of heart valves.
Depending on the conditions of the experiments, time
gain compensation can be used. This compensates for
attenuation of the signal as it travels deeper and deeper into
the imaged region. Generally, the details of the correction
function are manually specified by the operator, in order to
equalise the intensity across the image.
As a final step, the acquired image may need to be cor-
rected for the shape of the transducer. A typical example of
this so-called scan conversion is shown in Fig. 4. Here the raw
data B-mode data (i.e. an intensity line per transducer element)
is transformed to the corresponding wedge-shaped image that
represents the true physical field-of-view for the curved trans-
ducer that was used. Naturally, this process requires some sort
of interpolation process that leads to an non-uniform spatial
resolution and to distorted particle images.
Implicitly, it has been assumed that the emission and
receiving was done sequentially using individual elements.
In practice phased-arrays are used in more elaborate ways,
generally using groups of elements to record a single
(A)
(B)
(C)
(D)
Fig. 3 Schematic representation of ultrasound imaging: a Overview of geometry, b raw RF signal, c log-compressed intensity signal, and d
B-mode image. See text for details
Exp Fluids (2017) 58:3
1 3
Page 5 of 28 3
A-mode line. A prime example is focussing, as is also dem-
onstrated in Fig. 2: here five elements are used, triggered
with an electronic delay so that the wave fronts combine to
a focal region. The number of elements that is used deter-
mines the aperture. Naturally, this method of focussing pro-
vides more flexibility than a fixed, physical acoustic lens.
Note that such a lens is still used to focus the beam in the
elevational direction (as opposed to the azimuthal focus-
sing done by the phased-array). In Fig. 2 the focal location
in the elevational and azimuthal plane are identical, but this
is not required. Multiple focal points at various depths can
often be selected, but this comes at the cost of imaging rate,
as they are acquired sequentially. Other uses of a phased-
array are beam steering (i.e. tilting the beam axis), and a
variable line density (with the option to create more lines
than transducer elements by means of interpolation).
While the focus of this review is on velocimetry, it
should be pointed out that ultrasound imaging by itself can
also be a useful tool in fluid mechanics (beyond the obvi-
ous cardiovascular applications). An illustrative example is
shown in Fig. 5, where the deformation of a compliant tube
due to a traveling pressure wave is visualised using ech-
ography. This example, taken from Hickerson and Gharib
(
2006), demonstrates that imaging is possible without the
need for refractive index matching, which can be difficult
to achieve if also mechanical properties (e.g. Young’s mod-
ulus) of the wall need to be matched.
2.3 Reflection, refraction and scattering
Ultrasound is an archetypical wave phenomenon. The
(small) pressure fluctuations satisfy the wave equation and
Fig. 4 Raw B-mode data of
microbubbles in water (left) and
the corresponding scan-con-
verted (right) image Repro-
duced from Kim et al. (
2004a).
©Springer-Verlag 2004. Used
with permission
Fig. 5 A pressure wave trave-
ling through a compliant tube
(from left to right) is visualised
using two consecutive ultra-
sound images. The position of
the field-of-view and transducer
location is shown schemati-
cally Image reproduced from
Hickerson and Gharib (
2006).
©Cambridge University Press.
Used with permission