Passive Elastography

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/224242506

Passive Elastography: Shear-Wave


Tomography From Physiological-Noise
Correlation in Soft Tissues

Article in IEEE Transactions on Ultrasonics Ferroelectrics and Frequency Control · July 2011
DOI: 10.1109/TUFFC.2011.1920 · Source: IEEE Xplore

CITATIONS READS

33 142

6 authors, including:

Thomas Gallot Philippe Roux


Universidad de la República de Uruguay University Joseph Fourier - Grenoble 1
32 PUBLICATIONS 227 CITATIONS 377 PUBLICATIONS 5,926 CITATIONS

SEE PROFILE SEE PROFILE

Javier Brum Nicolás Benech


Universidad de la Republica, Montevideo, Ur… Universidad de la República de Uruguay
47 PUBLICATIONS 291 CITATIONS 37 PUBLICATIONS 593 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Metamaterial physics applied to seismic waves View project

Ocean Acoustic Tomography View project

All content following this page was uploaded by Philippe Roux on 09 September 2014.

The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Passive Elastography: Shear-Wave Tomography from Physiological-Noise Correlation
in Tissues

T. Gallot,1 S. Catheline,1 P. Roux,1 J. Brum,2 N. Benech,2 and C. Negreira2


1 Laboratoire de Géophysique Interne et Tectonophysique,
CNRS-Université J. Fourier, 38041 Grenoble, France
2 Laboratorio de Acústica Ultrasonora, Instituto de Física Facultad de Ciencias,
Universidad de la República Montevideo - Uruguay, 11400 Montevideo - Uruguay
Inspired by seismic-noise correlation and time reversal, a shear-wave tomography of soft tissues
using an ultrasonic scanner is presented here. Free from the need for controlled shear-wave sources,
this passive elastography is based on Green's function retrieval and takes advantage of the permanent
physiological noise of the human body. Results with hydrogels indicate that this tomography tech-
nique is compatible with low frame rates used by conventional echographic and magnetic-resonance
imaging devices. Furthermore, in-vivo experiments on the liver demonstrate the pertinence of ap-
plying tools of seismology in the eld of medical imaging. Passive elastography should now provide
benets for medical applications, whereby determination of local tissue elasticity can support and
improve diagnostic condence.
PACS numbers: pacs number...

Correlation of noise-like signals is a recent method for relies on the shear noise within the human body. Sec-
the extraction of information from multiply scattered ond, we demonstrate that the technique is compatible
wave elds [1]. The most important application to date with conventional low-frequency imaging devices.
relates to geophysics and crustal tomography obtained The feasibility of this correlation-based passive tomog-
from seismic-noise correlation [2, 3]. This consists of raphy is rst presented in a tissue-mimicking-phantom of
extracting a surface-wave velocity map from the low- polyvinyl alcohol (PVA gel; [16]). Contrary to seismol-
amplitude permanent ambient noise in the earth sub- ogy, where the seismic noise can only be recorded at or
surface. Passive imaging methods based on noise cor- close to the earcth surface, ultrasound in soft solids al-
relation have also been successfully developed in various lows the measurement of the particle velocity inside the
other domains and at various scales, such as in helioseis- volume along ultrasonic beams (z -axis in Fig. 1A). A 64-
mology [4, 5], oceanography [6, 7], and ultrasonics [8]. transducer array connected to a Lecoeur ultrafast scan-
Due to muscular activity, there is physiological acous- ner was used to measure the ultrasonic echoes from a ho-
tic noise in the body, and this can be used to extract mogeneous distribution of scatterers inside the PVA gel,
the mechanical properties of tissues. Recently, Sabra et with a repetition frequency of 1 kHz. Speckle interferom-
al. showed [9] that the natural vibrations of contracting etry algorithms allowed the tracking of the echoes and
muscles can provide measurements of their global vis- deduced the scatterer displacement between each shot.
coelastic properties. In the present study, passive in-vivo As proof of concept for passive tomography, a 9 s noise-
tomography of soft tissues is proposed, based on the cor- like elastic eld was created by nger impacts applied
relation of a physiological noise eld. This natural elastic randomly all over the surface (Fig. 1A). To give the
eld that is dominated by low-speed shear waves is per- same weight to each frequency, a whitening processing
manently induced by breathing and cardiac activities. It was applied between 10 Hz and 200 Hz. The displace-
can be measured at the body surface using ultrasonic ment ψz (0, t) along the rst transducer at 3 cm inside
techniques that have been developed in the eld of elas- the gel is shown in Figure 1B. The zoom of 0.5 s of the
tography, which are similar to Doppler methods [10, 11]. displacement eld along one line parallel to the array (x -
At present, an ultrasonic device in the MHz frequency axis) is shown as a color-scale map (Fig. 1C). The n-
band can record the motion of the ultrasonic speckle at a ger impacts generated a low-frequency random-like dis-
rate of 1,000 frames per second. This ultrafast imaging is placement eld that was dominated by shear waves that
needed to measure shear-wave propagation in the human clearly showed strong spatial coherence along the x-axis;
body, where the typical frequencies range from 10 Hz to e.g., around t = 1.4 s in Figure 1C.
300 Hz. Using this technology, the active methods that The cross-correlation C (x0 , x; t) processed at two
are dened as transient elastography take advantage of points x0 and x should naturally highlight the spatio-
an external source [12] or a radiation force [13] to clini- temporal coherence of the noise-induced displacement
cally characterize the elasticity of soft biological tissues eld, as in Eq. (1).
[14, 15]. Two improvements are proposed in the present
study: rst, the shear-wave source is removed and the C (x0 , x; t) = ψz (x0 , T − t) ⊗ ψz (x, t) (1)
method becomes purely passive, in the sense that it only
In the case of a spatially homogeneous distribution of
2

Figure 1: A) Experimental set-up for the in-vitro correlation-


based tomography. A complex shear-wave eld was created Figure 2: Correlation maps measured at six refocusing spots
by nger impacts applied to the surface of a tissue-mimicking x0 in a homogeneous hydrogel. The converging and diverging
phantom (PVA gel). Displacements were measured inside the waves are responsible for this typical cross-like shape of the
volume using an ultrafast ultrasonic scanner and elastography TR eld in this spatio-temporal representation. The group
algorithms. B) The z -component of the particle velocity at x velocity of these waves can be estimated from the delay δt of
= 0, z = 3 cm, the rst point of the red line. C) The zoom the maximum of the eld along δx (bottom right). Another
of 0.5 s of the particle velocity along the red line shows the approach estimates the -6 dB focal spot width close to the
wave propagation around 1.4 s. half wavelength λ/2(top right).

white noise sources [17], the noise-correlation function mation of the shear waves at each position,correlation-
C (x0 , x; t) can be shown to be related to the Green's based techniques take advantage of: 1) the spatial diver-
function G(x0 , x; t) between xo and x, according to Eq. sity of the noise sources; and 2) the long duration of the
(2): noise recording. In the following, two independent meth-
d ods were compared to measure the local shear velocity,
C(x0 , x; t) = G(x0 , x; −t) − G(x0 , x; t) (2) which in soft solids is directly related to the elasticity
dt
properties.
The analogy with time-reversal (TR) physics [18] follows Two PVA phantoms were prepared with dierent scat-
directly from Eq. (2), as the sum of the causal and terer concentrations. The stiness was independent of
acausal Green's function is exactly what would be ob- the scatterer concentrations [16], although it created an
served on x in a perfect TR eld with a source at x0 apparent interface on a sonogram (Fig. 3A). The sti-
sending a pulse at t = 0 s [19]. In this interpretation, the ness of this hydrogel was controlled by the number of
correlation function consists of using each noise source freeze-thaw cycles: one cycle for the soft gel on the top
excited by the nger impacts as part of a TR mirror that (0 < z < 15 e 20 mm), and ve cycles for the sti gel
refocuses on x0 [20]. This noise-correlation-based refo- below. Transient elastography [10] was used for the stan-
cusing is similar to the TR eld of elastic waves that were dard measurement of the shear velocities, as 4.3 m/s and
recently observed in a soft solid cavity [21]. The noise- 1.2 m/s respectively.
correlation function C(x0 , x; t) is shown for six refocusing In the rst method, the speed of the converging and
spots x0 (Fig. 2). For each position, the amplitude was diverging shear waves were measured. According to elas-
maximal at x = x0 and t = 0, as predicted by Eq. (2). ticity theory, the speed is the key parameter, as the shear-
The classical cross-shape of a focused eld was seen along wave speed cs is related to the shear elasticity µ and also
the x-axis, showing the converging wave for t < 0, the µ 3E
refocusing at t = 0, and the diverging wave for t > 0. to the Young's modulus E , according to : c2s = ≈ .
ρ ρ
The TR eld emerges on each spot without the need for The whole frequency spectrum of the refocusing was used
any extra spatial averaging. Compared to the equivalent to estimate the local group velocity. Spatial ltering was
TR experiment using a unique point-like source in soft applied to the correlation eld to select only one direc-
solids [21], a signicant improvement in signal-to-noise- tion of propagation (e.g., along the x-positive direction).
ratio is obtained in this correlation-based experiment, The group velocity at frequencies ranging from 10 Hz to
where many nger impacts can create a quasi-random 100 Hz were deduced from the delays δt between two con-
noise-like eld. Indeed, to provide the local speed esti- secutive local eld maxima separated by δx = 0.8 mm.
3

For the sake of clarity, δx was set at 18 mm for Figure →



r , as in Eq. (3):
2 (bottom right panel). This technique was applied to ˆ T
construct the shear-wave-speed tomography of the two- C (→

r0 , →

r ; 0) = ψz (→

r0 , τ ) .ψz (→

r , τ ) dτ (3)
layered hydrogel. The velocity along the x-axis was es- 0
timated for 83 lines at successive depths, resulting in a
49×49 mm² stiness map. The shear velocity is shown as To benet from the self-averaging properties of the TR
a color-scale map (Fig. 3B). The averaged shear velocity process [24], the refocusing was computed using a large
in the two media, at 4.8 m/s and 1.7 m/s, were close to frequency bandwidth. In this bandwidth, the frequency
the measurements made by standard transient elastog- associated with the estimated wavelength was the up-
raphy [10]. Furthermore, the interface between the two per frequency fmax [25]. Consequently, for an easier
media that was seen from the ultrasonic echoes in Figure comparison, the wavelength tomography was translated
3a closely matched the shear-velocity map (Fig. 3B). into speed tomography (Fig. 3B), using the relationship
v =λf .
Basically, the wavelength tomography of this second
method was independent of the speed tomography of the
rst method. However, opportunely, they coincided (Fig.
3B and C). In both of the layers, the average wave speed
was 4.1 m/s and 1.7 m/s for the sti and the soft gels,
respectively, which was in good agreement with the classi-
cal active-transient elastography measurements (4.3 m/s
and 1.2 m/s). The wavelength approach, however, was
more robust with regard to the signal-to-noise ratio [25],
as the wavelength was estimated when the intensity was
maximal.
A second advantage of this wavelength tomography is
its compatibility with slow imaging devices that cannot
reach the typical 1 kHz sampling frequency of an ultrafast
scanner. Under the Shannon frequency (< 400 Hz), the
temporal coherence of the experimental elastic eld was
progressively lost, as the maximum observed frequency
was around 200 Hz. However, in the worst case, despite
a total loss of time coherence, the spatial coherence was
conserved and the refocusing map can still be used for
wavelength estimation. Mathematically, this loss of time
Figure 3: A) Sonogram of the bilayer hydrogel, where the coherence can be written according to Eq. (4), by adding
two layers with dierent scatterer concentrations are clearly a random phase φ to the time argument of the displace-
apparent. B) Shear-wave-speed tomography estimated ment ψz :
from local group-speed estimation. C) Shear-wavelength ˆ
tomography estimated from the refocusing width. D) Slow
T

shear-wavelength tomography estimated from a displace- Cφ (→



r0 , →

r ; 0) = ψz (→

r0 , φ (τ )) .ψz (→

r , φ (τ )) dτ (4)
0
ment eld sampled at 50 Hz. The interface between the two
layers is in the same position in the four images, and good Due to the time integration, this phase does not mod-
quantitative agreement was found for the shear-speed values. ify the correlation map at t = 0, C (→ −
r0 , →

r ; 0) =

− →

Cφ ( r0 , r ; 0). This means that the estimation of wave-
length tomography is feasible, despite the slow imaging
In a second method, the shear-wave speed was de- frequency. However, the loss in the frequency information
duced from estimation of both the wavelength and the prevents estimation of the maximum frequency, and the
frequency, through the relationship v = λ f . In a TR shear speed cannot be deduced directly. Thus, only the
eld, the Rayleigh criteria stipulate that the focal spot shear wavelength was used to construct the tomography
width is close to the half wavelength [22]. Consequently, shown in Figure 3D. In this experimental demonstration,
the -6 dB width measured for each position → −
r0 of Fig- a 50-Hz sampling frequency was used. The interface was
ure 2 (top right panel), gives the wavelength estimation clearly apparent, with the contrast between the two lay-
directly. This is known as the spatial auto-correlation ers similar to that seen for Figure 3B and C. To relate
method in seismology [23], and it is based on the correla- the measured wavelength to the shear speed, the maxi-
tion function C (→−
r0 , →

r ; 0) at the refocusing time (t = 0) mum frequency was estimated from the raw displacement
between seismic stations, or in the present study, between eld sampled at 1 kHz. On a conventional ultrasonic de-
the refocusing point → −
r0 and the 64 × 83 acquisition points vice, this estimation can be performed in a limited region
4

of interest, as for the Doppler mode. The maximum fre- to the abdominal muscles, whereas the dark region was
quency was around fmax '50 Hz, which implied averaged the liver. Good correspondence was seen with the shear-
shear-wave speeds of 3.7 m/s and 1.9 m/s in the hard and wave-speed tomography (Fig. 5B), where the interface
soft gels, respectively, which were close to the expected between these two regions was seen around z ' 12mm. In
values. muscle and liver, the averaged shear-wave speeds were in
In these phantom experiments, the noise-like elastic quantitative agreement with other studies [26, 27]. The
eld was created by nger impacts. For the in-vivo exper- variability of the reconstructed tomographies shown in
iment, the physiologic noise that is permanently present Figures 3 and 5 are believed to be related to the direc-
in the body creates a natural complex shear-wave eld. tivity variance of the shear-wave eld. Indeed, it is well
Thus the liver was imaged at a 750-Hz repetition fre- established in seismology that the ux resulting from a
quency, by placing the transducer array under the ribs diuse eld [28, 29] can be responsible for bias in the to-
of the subject (Fig. 4A). In Figure 4B, the particle dis- mography. Special eorts need to be made to compensate
placements in the low-frequency bandwidth from 5 Hz to for this eect.
70 Hz are mainly due to cardiac activity. The displace-
ment eld in the liver shows a cardiac pulse each 0.5 s,
followed by reverberation and scattering. This complex
eld was used to compute the correlation function (Fig.
4C).

Figure 5: A) Sonogram of the liver region. The interface


between the abdominal muscles and the liver is visible around
z = 12 mm. B) The passive shear-wave-speed tomography
from the correlation width clearly shows the two regions. The
averaged shear-speed estimations are in agreement with values
in the literature.

In summary, while other stiness-tomography tech-


Figure 4: A) Experimental set-up for the in-vivo correlation- niques actually suer from the physiological noise in
based tomography: an ultrafast scanner was used to measure
the natural displacement eld in the liver region. B) The par- the human body, the proposed method takes advantage
ticle velocity along an acquisition line at 4 cm, and parallel to of this permanent elastic eld as passive elastography,
the array, shows the physiological elastic eld. C) In the cor- in the sense that no articial shear-wave sources were
relation map C(x0 , x; t) with x0 = 14 mm, only one direction needed. Future studies will focus on the heart, where
of propagation emerged from the refocusing eld. elastography with active methods is only achievable with
diculty. Passive elastography is also compatible with
Compared to the refocusing eld in Figure 2, only one low-frequency acquisition systems, such as standard two-
branch of the cross-shape was recovered, which can be dimensional and three-dimensional ultrasound and mag-
explained by the strong directivity of the eld. In other netic resonance.
words, using a TR analogy, we were not dealing here with
a perfect TR cavity, where the converging wave was com-
ing from all directions of a closed surface, but rather with
a TR mirror that was located around the heart region and [1] E. Larose et al. Geophysics, 71, SI11-SI21 (2006).
that was re-emitting waves in the direction of the liver. [2] M. Campillo, A. Paul, Science, 299, 547-549 (2003).
In addition to this problem of directivity, it should be [3] N.Shapiro, M. Campillo, L. Stehly, M. H. Ritzwoller, Sci-
noted that the particle velocity amplitudes were weak, ence, 307, 1615-1618 (2005).
at only 7.5 mm/s, which was quite close to the sensi- [4] T. L. Duvall Jr, S. M. Jeeries, J. W. Harvey, M. A.
tivity limit of the elastography, at around 1 mm/s. For Pomerantz, Nature, 362, 430-432 (1993).
these two reasons, only the more robust technique, i.e., [5] P. M. Giles, T. L. Duvall Jr, P. H. Scherrer, R. S. Bogart,
the wavelength approach, was used for in-vivo passive Nature, 390(6655), 52-54 (1997).
[6] M. J. Buckingham, B. V. Berkhouse, and S. A. L. Glegg,
tomography. Nature (London), 365, 327-329 (1992).
In the sonogram data (Fig. 5A), two zones were visi- [7] P. Roux, W. A. Kuperman, J. Acoust. Soc. Am. 116,
ble. Above z ' 12 mm, the bright region corresponded 1995-2003 (2004).
5

[8] O. I. Lobkis, R. L. Weaver, J. Acoust. Soc. Am. 110, electr. 3(5), 579-592 (1992).
3011-3017 (2001). [20] A. Derode, E. Larose, M. Campillo, and M. Fink. Appl.
[9] K. Sabra, S. Conti, P. Roux, W Kuperman. Appl. Phys. Phys. Lett. 83(15), 3054-3056 (2003).
Lett. 90, 194-101, (2007). [21] S. Catheline, N. Benech, J. Brum, and C. Negreira Phys.
[10] S. Catheline, F. Wu, M. Fink, J. Acoust. Soc. Am. Rev. Lett. 100, 064301-4 (2008).
105(5), 2941-2950 (1999). [22] J. De Rosny, M. Fink, Phys. Rev. Lett. 89, 124301-4
[11] J. Ophir et al. Ultrason. Imaging, 13(2), 111-134 (1991). (2002).
[12] L. Sandrin et al. IEEE Trans. Ultrason. Ferroelectr. [23] K. Aki, Bull. Earthquake Res. Ins. 35, 415-456 (1957).
49(4), 426-435 (2002). [24] G. Papanicolaou, L. RYZHIK, K. SØLNA, SIAM J.
[13] J. Berco, M.Tanter, M. Fink, IEEE Trans. Ultrason. Appl. Math. 64(4), 1133-1155 (2004).
Ferroelectr. 51, 396-409 (2004). [25] N. Benech et al. IEEE Trans. Ultrason. Ferroelectr.
[14] Y. Yamakoshi, J. Sato, T. Sato, IEEE Trans. Ultrason. 56(11), 2400-2410 (2009).
Ferroelectr. 37(2), 45-53 (1990). [26] J.L. Gennisson et. al. j. biomech. 38(7), 1543-1550
[15] K. J. Parker, SR Huang, RA Musulin, RM Lerner, Ul- (2005).
trasound Med. Biol. 16(3), 241-246 (1990). [27] M. Muller, et al. Ultrasound in Med. & Biol. 35(2), 219-
[16] J. Fromageau, et al. IEEE Trans. Ultrason. Ferroelectr. 229 (2009).
54(3), 498-508 (2007). [28] R. Weaver, B. Froment, M. Campillo, J. Acoust. Soc.
[17] P. Gouedard et al. Geophysical Prospecting, 56(3), 375- Am. 126(4), 1817-26 (2009).
393 (2008). [29] P. Gouedard, P. Roux, M. Campillo, A. Verdel. Geo-
[18] M. Fink, Phys. Today, 34, 34-40 March (1997). physics, 73(6), V47-V53 (2008).
[19] D. Cassereau, M. Fink, IEEE Trans. Ultrason. Ferro-

View publication stats

You might also like