IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, . 59, . 11, NOVEMBER 2012 2390
0885–3010/$25.00
©
2012 IEEE
The Variance of Quantitative Estimates
in Shear Wave Imaging:
Theory and Experiments
Thomas Deffieux, Jean-Luc Gennisson, Benoit Larrat, Mathias Fink, and Mickael Tanter
Abstract—In this paper, we investigate the relationship be-
tween the estimated shear modulus produced in shear wave
imaging and the acquisition parameters. Using the framework
of estimation theory and the Cramer–Rao lower bound applied
both to the estimation of the velocity field variance and to the
estimation of the shear wave travel time, we can derive the
analytical formulation of the shear modulus variance σ
μ
2
using
relevant physical parameters such as the shear wave frequency,
bandwidth, and ultrasonic parameters. This variance corre-
sponds to the reproducibility of shear modulus reconstruction
for a deterministic, quasi-homogeneous, and purely elastic me-
dium. We thus consider the shear wave propagation as a deter-
ministic process which is then corrupted during its observation
by electronic noise and speckle decorrelation caused by shear-
ing. A good correlation was found between analytical, numeri-
cal, and experimental results, which indicates that this formu-
lation is well suited to understand the parameters’ influence in
those cases. The analytical formula stresses the importance of
high-frequency and wideband shear waves for good estimation.
Stiffer media are more difficult to assess reliably with identical
acquisition signal-to-noise ratios, and a tradeoff between the
reconstruction resolution of the shear modulus maps and the
shear modulus variance is demonstrated. We then propose to
use this formulation as a physical ground for a pixel-based
quality measure that could be helpful for improving the recon-
struction of real-time shear modulus maps for clinical applica-
tions.
I. I
T
recent development of quantitative elastography
methods is regarded as a highly promising step toward
improved diagnosis. Since its introduction, many studies
have been published by several groups demonstrating sev-
eral clinical applications such as breast cancer detection
[1]–[4], liver fibrosis staging [5]–[9], kidney monitoring
[10], thyroid gland [11] and prostate cancer detection [12],
and musculoskeletal monitoring [13]–[15], ophthalmologic
[16], cardiac [17], or vascular applications [18]. Some clini-
cal applications, such as liver fibrosis staging [19], require
a very accurate and quantitative estimation of the shear
modulus to make a diagnosis because the estimated value
is directly compared with a pre-established quantitative
scale. A quantitative approach is also mandatory for the
longitudinal monitoring of drug treatment efficacy (such
as chemotherapy, antiangiogenic, or antivascular treat-
ments).
To allow a quantitative estimation of the shear mod-
ulus of tissue, dynamic elastography techniques—or in
other words, shear-wave-based imaging techniques—use
mechanical shear waves whose propagation velocity is di-
rectly linked to the shear modulus of the medium, i.e., its
stiffness. After generation of shear waves in the medium,
their propagation is recorded either in real time or using
a stroboscopic acquisition. By solving the wave equation,
it is then possible to retrieve a quantitative value of the
shear modulus. There are generally two groups of dynamic
elastography techniques depending on the regime of the
shear waves used. In steady-state elastography, such as
in magnetic resonance (MR) elastography, a monochro-
matic shear wave is generated and imaged with a strobo-
scopic acquisition, but in 3-D, using an MR imaging sys-
tem. These techniques were first developed by Muthupillai
and Manduca [20], [21] and are used today for clinical
research. Although they allow the full 3-D reconstruction
of the shear wave field, the overall complex patterns of the
steady-state shear waves require the complete solving of
the shear wave equation, which leads to difficulties with
regard to robustness and bias because the estimation of
second-order derivatives is required [22], [23]. Sonoelastic-
ity [24] is another example of a steady-state elastography
technique, in which shear waves are acquired with an ul-
trasound system.
Conversely, in transient elastography, such as in 1-D
transient elastography (TE) [25], shear wave elasticity im-
aging (SWEI) [26], acoustic radiation force imaging with
shear wave speed estimation (ARFI SWS) [27], supersonic
shear imaging (SSI) [28], shear wave dispersion ultrasound
vibrometrey (SDUV) [29], spatially modulated ultrasound
radiation force (SMURF) [30], dynamic micro elastog-
raphy (DME) [31] or transient MR elastography [32] (t-
MRE), a transient—and consequently wideband—shear
wave is generated and tracked during its propagation ei-
ther in real time or using a gated acquisition. The key dif-
ference is that the shear wave is imaged before it can reach
the boundaries and is generally considered as a transient
wave front. Under this assumption, the shear modulus can
then be estimated by solving the eikonal equation for the
wave front instead of the full wave equation with much
simpler algorithms, such as a time-of-flight algorithm [2]
which locally estimates the shear group velocity along the
propagation direction. This approach is more robust to
Manuscript received November 18, 2011; accepted July 6, 2012. This
work was partly supported by the French national agency for research on
AIDS and Viral Hepatitis (ANRS).
The authors are with the Institut Langevin, Ondes et Images, Ecole
Supérieure de Physique et de Chemie Industrielles (ESPCI, ParisTech),
Centre National de la Recherche Scientifique (CNRS) UMR 7587,
INSERM U979, France (e-mail: tdeffieux@gmail.com).
DOI http://dx.doi.org/10.1109/TUFFC.2012.2472