SVD beamforming for ultrafast aberration correction

and real-time speed-of-sound quantification

Hanna Bendjador1, Thomas Deffieux1, Mickaël Tanter1
1Physics for Medicine Paris, Inserm, ESPCI Paris, CNRS, PSL University
Paris, France
hanna.bendjador@gmail.com

Abstract— The rise of plane or diverging ultrasound waves Essentially, they rely either on Phase Aberration Correction
allowed to perform high quality imaging at much faster frame (PAC)[10][4][11] or on coherence optimization [12][13]. In
rates. Though, propagation through complex layers induces this work, we propose a novel and fast aberration correction
phase and amplitude aberrations that remain partially approach in the context of ultrafast ultrasound imaging, based
unresolved. In ultrafast ultrasound imaging, the backscattered on ultrafast coherence optimization. We demonstrate that it
signals contain information on the whole medium, for each allows in real time, both adaptive imaging and Speed-Of-
transmission. We proposed an approach to take benefit of this Sound (SoS) quantification. Strikingly, it combines in a single
and provide a fast aberration correction. Based on the Singular mathematical operation, the phase and amplitude aberration
Value Decomposition of an appropriate matrix (the ultrafast
correction with the coherence optimization of backscattered
compound matrix), this technique separates the spatial and
angular variation. We demonstrate that the first eigen vector
data.
optimizes the angular coherence of backscattered signals. Thus, II. THEORY
it provides the knowledge of the amplitude and phase of the
aberration on one side, and the corrected image on the other A. The Ultrafast Compound Matrix
side. Experimental in vivo results prove the efficiency of the
We consider the transmission of 𝑁𝜃 plane waves through
image correction, even in the case of spatially extended
aberrations. Thanks to the knowledge of the aberration phase,
the medium. We introduce a new matrix: the Ultrafast
we proposed a real-time estimation of the speed of sound within Compound Matrix 𝑅. It contains the plane wave images at
the region of interest. We implemented this further to build local each transmitted angle, beamformed both in emission and
in vitro sound speed maps. The speed and simplicity of this novel reception. The summation of these images leads to the
approach paves the way to unprecedented quantitative and classical ultrafast compound image [14]. We chose to keep all
adaptive ultrafast ultrasound applications. the information of plane wave transmits in the single matrix
R, of size the number of pixels in x and z (𝑁𝑥 , 𝑁𝑧 ) times the
Keywords— Adaptive Beamforming, Ultrafast Imaging, number of plane waves (𝑁𝜃 ). It is reshaped in a 2D-Casorati
Singular Value Decomposition, Aberration Correction. matrix, with all the pixels on one dimension and all the
transmit events on the other.
I. INTRODUCTION
Ultrasound imaging consists in an inverse problem
between the echogenicity image of a medium, and its
backscattered echoes on a transducer array. In most cases, the
Delay-And-Sum method is very efficient to solve this
problem. Ultrafast ultrasound introduced plane waves to form
high quality images at high frame rates. This gave rise to
several ultrasound modalities such as Shear Wave
Elastography, Functional Neuroimaging or Ultrafast Doppler.
Still, in complex media such as fat, muscle or bone layers, the
sound speed inhomogeneities induce wave front distortions
altering both the image quality, and the quantitative
assessments [1][2]. Thus, two challenges remain to fully
compensate acoustic aberrations:
- Correcting the images through strongly aberrated
layers.
- Quantifying the sound speed [3][4], a novel and Figure 1: Construction of the Ultrafast Compound Matrix R. Plane
effective biomarker of diseases such as Non waves are beamformed individually. Their sum is the classical
Alcoholic SteatoHepatitis (NASH) [5]. ultrafast compound image. The matrix R is their concatenation,
Some techniques have been developed to tackle these reshaped to form a 2D-matrix.
issues, such as speckle brightness [6], time reversal of speckle
noise [7], or computed ultrasound tomography [8][9].

978-1-7281-5448-0/20/$31.00 ©2020 IEEE

Authorized licensed use limited to: Cornell University Library. Downloaded on May 24,2021 at 09:34:40 UTC from IEEE Xplore. Restrictions apply.
 III. METHODS
B. Aberration and angular covariance A. The SVD Beamformer approach
In a given region of interest, we define a reduced version Ultrasound acquisitions were performed on a
of 𝑅: 𝑅̃ . In order to optimally correct ultrasound images, we programmable research scanner (Verasonics, Vantage 256)
intend to maximize the coherence of backscattered echoes driving a 192-element probe (central frequency: 6.25 MHz,
between plane wave transmits. We define the Local Angular pitch: 0.2 mm). Plane wave sequences – from 5 to 100
Coherence as the covariance matrix of a restricted version transmits in [-18°, 18°] - were designed with a repetition
of 𝑅: frequency of 10 kHz. We tested in vitro gelatin phantoms of
𝑡 ̃∗ different fat contents, in vivo human liver and in vivo
𝐶𝜃 = 𝑅 . 𝑅̃ (1) transcranial mouse brain.
In speckle, the angular covariance decreases linearly with
the angle lag: it is the Van Cittert Zernike theorem adapted for
angular coherence [15]. When an aberrating layer is placed
between the probe and the medium, the angular coherence
sinks dramatically. Thus, its optimization is one of the keys
for aberration correction.

Figure 3: Priniciple of the SVD Beamforming algorithm. Singular Value

Decomposition is performed on the R matrix.

As mentioned earlier, the Singular Value Decomposition

of 𝑅 provides, in its first angular eigen vector, the amplitude
and phase aberration law in the plane wave basis [6]. The first
spatial eigen vector provides directly the corrected image.

B. The case of spatially extended aberrations

Figure 2: Amplitude of the angular covariance matrix (left), and In many experimental configurations, the phase and
coherence factor (right). (a): Ordinary speckle. (b) Aberrated amplitude screen approximation is no longer valid to model
speckle. the aberration. To adress them, we propose to perform the
Let us note A the angular aberration correction matrix in the SVD on isoplanatic patchs, within which the aberration
phase and amplitude screen aberration approximation. The remains consistent.
matrix RA is then the corrected image matrix. So, its angular
covariance 𝑡(𝑅𝐴)∗ (𝑅𝐴) is optimal. Now, this can be written
as a Rayleigh quotient maximization:
𝑡 𝑡 ∗ 𝑡 ∗
(𝑅𝑋)∗ . (𝑅𝑋) 𝑋 ( 𝑅 𝑅) 𝑋
𝐽(𝑋) = ∗
=
𝑋 .𝑋 ‖𝑋‖

Mathematically, its solution is given by 𝑋 being the first

eigen vector of 𝑡𝑅 ∗ 𝑅 . This matrix being Hermitian, the
previous assertion is equivalent to 𝑋 being the first singular
vector of 𝑅. So, the angular aberration correction matrix 𝐴 is
the first singular vector of 𝑅.

This work was supported by the Inserm ART (Technology Research Figure 4: Result of SVD Beamforming on different patches within
Accelerator) "Biomedical Ultrasound", the European Research Council a three-layer phantom. (a) B-mode image displaying three patches.
(ERC Advanced Grant FUSIMAGINE) and the Fondation pour la Recherche (b) Corresponding phase aberration determined from the first
Médicale (F.R.M.). angular singular vector

Authorized licensed use limited to: Cornell University Library. Downloaded on May 24,2021 at 09:34:40 UTC from IEEE Xplore. Restrictions apply.
 Also, we designed a custom phantom, made of cubes of
This gives us access to an individual complex aberration for gelatin-agar phantoms at different sound speeds. We
each patch, and allows to solve strong aberrators, with much performed the SVD Beamformer on isoplanetic patches,
more complex descriptions than the screen approximation. On center in each pixel of the image. Fig. 7 demonstrates the
the particular example of sound speed phantom layers of Fig. ability of our technique, after a simple derivative approach, to
4, we also observe a change of curvature in each layer, that compute the local sound speed map with a deviation better
predicts our ability to deduce a local sound speed estimate. than 1% of expected values.
IV. RESULTS
A. The first spatial eigen vector
On pin and anechoic cyst phantoms, we demonstrated in
previous works that the SVD Beamformer corrects
successfully aberrations, both in contrast and resolution [16].
We decided to apply our approach to an in vivo example of
distributed aberration: the transcranial brain imaging of the
mouse. On B-mode (a), Fig. 5, the images suffer a lot from the
distortion of the wavefront induced by the heterogeneity of the
skull bone.

Figure 7: Results of sound speed estimation on in vitro custom

phantom: (a) B-mode image. (b) Computed sound speed map.
(c) Resume of mean values estimated and expected in each
Figure 5: B-mode images of a mouse brain through the skull. (a) Raw section of the medium.
image. (b) Image obtained after correction with SVD Beamforming.

After SVD Beamforming, B-mode images showed a huge C. Performance on ultrafast data
improvement in contrast: +8.1 dB. The interfaces definition
qualitatively looks thinner after correction, eventhough To evaluate the performance of the SVD Beamformer on
structures inside the brain are not yet clearly separable. ultrafast applications, we assessed its efficiency for a low
Strikingly, this correction is already computed fast enough to
amount of angles. We evaluated contrast and resolution
allow real-time adaptive imaging.
improvements on in vitro phantoms (Fig. 8). Interestingly, the
correcting effect remains remarkable for as low as five
transmit angles. The method provides an important correction
B. The first angular eigen vector
of aberrated images, getting closer to the ideal image that
The knowledge of the phase and amplitude aberration was would have been obtained without any aberration.
used to fit propagation models, and thus, estimate in real time
the integral sound speed along the path. We verified this on in
vivo liver data. The link between sound speed and fat content
described by [17] grants immediate access to the fat fraction
quantification. Fig. 6 shows liver images of two healthy
volunteers. We provide, for each, a percentage of fat within a
region of interest. We excluded strong reflectors to avoid
artifacts in singular value decomposition of the images.

Figure 6: Liver B-mode images of healthy volunteers, showing

speed-of-sound and fat fraction estimation in a region of
interest. (a) Volunteer 1. (b) Volunteer 2.

Authorized licensed use limited to: Cornell University Library. Downloaded on May 24,2021 at 09:34:40 UTC from IEEE Xplore. Restrictions apply.
 ultrasound-based hepatic steatosis assessment,”
Phys. Med. Biol., vol. 62, no. 9, pp. 3582–3598, 2017.
[6] L. Nock, G. E. Trahey, and S. W. Smith, “Phase
aberration correction in medical ultrasound using
speckle brightness as a quality factor.,” J. Acoust.
Soc. Am., vol. 85, no. 5, pp. 1819–33, 1989.
[7] G. Montaldo, M. Tanter, and M. Fink, “Time reversal
of speckle noise,” Phys. Rev. Lett., vol. 106, no. 5, pp.
1–4, 2011.
[8] P. Stähli, M. Kuriakose, M. Frenz, and M. Jaeger,
“Forward model for quantitative pulse-echo speed-
of-sound imaging,” arxiv, vol. 1, no. 1902.10639v2,
pp. 1–22, 2019.
[9] M. Jaeger, G. Held, S. Peeters, S. Preisser, M.
Grünig, and M. Frenz, “Computed Ultrasound
Figure 8: Effects of SVD Beamformer as a function of the number
Tomography in Echo Mode for Imaging Speed of
of transmitted plane waves. (a) Comparison of aberrated and Sound Using Pulse-Echo Sonography: Proof of
corrected images for 100 angles. (b) Same comparison for 5 Principle,” Ultrasound Med. Biol., vol. 41, no. 1, pp.
angles. (c) Resolution plots depending on number of transmitted 235–250, 2015.
plane waves; comparison between no aberration, aberration, [10] R. Mallart and M. Fink, “Adaptive focusing in
and after correction. (d) Contrast plots in the same configuration scattering media through sound‐speed
as (c). inhomogeneities: The van Cittert Zernike approach
and focusing criterion,” J. Acoust. Soc. Am., vol. 96,
no. 6, pp. 3721–3732, 1994.
CONCLUSION
[11] G. Chau, M. Jakovljevic, R. Lavarello, and J. Dahl,
Finally, we presented a novel and fast method for “A Locally Adaptive Phase Aberration Correction
real-time adaptive imaging, and speed of sound estimation. (LAPAC) Method for Synthetic Aperture
Interestingly, we combine Phase Aberration Correction Sequences,” Ultrason. Imaging, vol. 41, no. 1, pp. 3–
techniques with Coherence Optimization approaches in the 16, 2019.
single operation of SVD of our designed Ultrafast Compound [12] J. Dahl, “Coherence beamforming and its
Matrix. A deeper interest in global inversion of the phase applications to the difficult-to-image patient,” IEEE
aberration will enable in the future precise speed of sound Int. Ultrason. Symp. IUS, 2017.
mapping. Implementation on ultrafast acquisitions will also [13] Y. Zhang, Y. Guo, and W. N. Lee, “Ultrafast
pave the way to novel adaptive applications, particularly Ultrasound Imaging Using Combined Transmissions
promising in transcranial brain imaging. with Cross-Coherence-Based Reconstruction,” IEEE
Trans. Med. Imaging, vol. 37, no. 2, pp. 337–348,
2018.
REFERENCES
[14] M. Tanter, J. Bercoff, L. Sandrin, and M. Fink,
“Ultrafast Compound Imaging for 2-D Motion
[1] S. Smith, G. E. Trahey, and O. T. von Ramm, “Two- Vector Estimation: Application to Transient
dimensional arrays for medical ultrasound,” Elastography,” IEEE Trans Ultrason Ferroelectr
Ultrason. Imaging, vol. 14, no. 3, pp. 213–233, 1992. Freq Control., vol. 49, no. 10, pp. 1363–1374, 2002.
[2] S. W. Flax and M. O’Donnell, “Phase-Aberration [15] Y. L. Li and J. J. Dahl, “Angular coherence in
Correction Using Signals From Point Reflectors and ultrasound imaging: Theory and applications,” J.
Diffuse Scatterers: Basic Principles,” IEEE Trans. Acoust. Soc. Am., vol. 141, no. 3, pp. 1582–1594,
Ultrason. Ferroelectr. Freq. Control, vol. 35, no. 6, 2017.
pp. 758–767, 1988. [16] H. Bendjador, T. Deffieux, and M. Tanter, “The SVD
[3] M. Jakovljevic, S. Hsieh, R. Ali, G. Chau Loo Kung, Beamformer: Physical Principles and Application to
D. Hyun, and J. J. Dahl, “Local speed of sound Ultrafast Adaptive Ultrasound,” IEEE Trans. Med.
estimation in tissue using pulse-echo ultrasound: Imaging, vol. X, no. X, p.
Model-based approach,” J. Acoust. Soc. Am., vol. 10.1109/tmi.2020.2986830, 2020.
144, no. 1, pp. 254–266, 2018. [17] M. Imbault et al., “Ultrasonic fat fraction
[4] M. A. Lediju, G. E. Trahey, B. C. Byram, and J. J. quantification using in vivo adaptive sound speed
Dahl, “Short-lag spatial coherence of backscattered estimation,” Phys. Med. Biol., vol. 63, no. 21, 2018.
echoes: Imaging characteristics,” IEEE Trans.
Ultrason. Ferroelectr. Freq. Control, vol. 58, no. 7,
pp. 1377–1388, 2011.
[5] M. Imbault et al., “Robust sound speed estimation for

Authorized licensed use limited to: Cornell University Library. Downloaded on May 24,2021 at 09:34:40 UTC from IEEE Xplore. Restrictions apply.