Ultrasound Elasticity Imaging Based Multilevel Estimation Using Radiofrequency Data

Abstract

An ultrasound elastography introduced to differentiate hard tumor inclusion embedded in soft tissue background based on similarity measurement of before and after deformation. In this study, freehand elastography has considered to localize hard inclusion embedded in soft tissue of phantom breast, where deformation generated by applying gentile compression using probe physical surface of ultrasound machine. Radiofrequency data of before and after deformation acquired and then processed off-line. A non-ability of refinement operation to regularize displacement estimation outliers at correlation window length of 2λ is addressed, where a multilevel processing algorithm has proposed to reinforce refinement operation by producing smooth elastography. In the first level of the processing, displacement field has estimated at correlation window length of 3λ, where global stretching as re-correlation operation and refinement operation as spatial regulation are included. While at second level, production of displacement estimation outliers at correlation window length of 2λ are regularized based on replacement of estimated cells with that interpolated one at first level. Results show an ability of multilevel algorithm to cope the issues that encountered previously proposed algorithm of refinement on estimation outlier free displacement field at an axial resolution of 2λ, and produces differential strain field.

Full text

IOP Conference Series: Materials Science and Engineering
PAPER • OPEN ACCESS
Ultrasound Elasticity Imaging Based Multilevel Estimation Using
Radiofrequency Data
To cite this article: Ali R.H. Al-jader et al 2020 IOP Conf. Ser.: Mater. Sci. Eng. 745 012105
View the article online for updates and enhancements.
This content was downloaded from IP address 185.95.207.93 on 21/03/2020 at 08:50

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd
The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
1
Ultrasound Elasticity Imaging Based Multilevel Estimation
Using Radiofrequency Data
Ali R.H. Al-jader1, Aws Alazawi2, Mazin N. Farhan3
1Department of Medical Instrumentation Techniques Engineering, Electrical Engineering
Technical College, Middle Technical University, Baghdad, Iraq
2Department of Medical Instrumentation Techniques Engineering, Electrical Engineering
Technical College, Middle Technical University, Baghdad, Iraq
3Department of Medical Instrumentation Engineering Techniques, Northern Technical
University
engalirakan3@gmail.com , aws_basil@mtu.edu.iq, mazin.nadheer@ntu.edu.iq
Abstract An ultrasound elastography introduced to differentiate hard tumor inclusion
embedded in soft tissue background based on similarity measurement of before and after
deformation. In this study, freehand elastography has considered to localize hard inclusion
embedded in soft tissue of phantom breast, where deformation generated by applying gentile
compression using probe physical surface of ultrasound machine. Radiofrequency data of
before and after deformation acquired and then processed off-line. A non-ability of
refinement operation to regularize displacement estimation outliers at correlation window
length of 2λ is addressed, where a multilevel processing algorithm has proposed to reinforce
refinement operation by producing smooth elastography. In the first level of the processing,
displacement field has estimated at correlation window length of 3λ, where global stretching
as re-correlation operation and refinement operation as spatial regulation are included. While
at second level, production of displacement estimation outliers at correlation window length
of 2λ are regularized based on replacement of estimated cells with that interpolated one at
first level. Results show an ability of multilevel algorithm to cope the issues that encountered
previously proposed algorithm of refinement on estimation outlier free displacement field at
an axial resolution of 2λ, and produces differential strain field.
Keyword: Breast tumor differentiation, sonography, sonoelastography, signal
processing, ultrasound radiofrequency data.
1. Introduction
The female breast cancer is one of the main reasons of death [1]. Medical imaging modalities of
ultrasound, mammography, and Magnetic Resonance Imaging (MRI), were routinely used for breast
cancer diagnosis. In particular, the ultrasound considered as non-ionize and non-invasive with a
capability to differentiate soft tissues [2], in addition to portability and low cost compared to
mammography and MRI. Ultrasound sonography and elastography widely used in medical
application to visualize tissue region of interest. Sonography used for abnormalities in biological
tissue[3], while ultrasound elastography introduced to differentiate hard inclusion embedded in
background soft tissues, based on mechanical characteristic of elasticity [4]. Elasticity determined by
estimation of displacement field using similarity measurement of cross correlation between windows

The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
2
of before and after applied compression. The low similarity between both windows of radiofrequency
data causes estimation outliers that produces non-differentiable strain field. Studies show that
decorrelation noise produced due to a reduction in similarity between reference and comparison
frames [5]. Re-correlation techniques were introduced to increase similarity between the frames, such
as Two Dimensional (2D) companding [6], Three Dimensional (3D) companding [7], adaptive
stretching [8]. Further studies considered coarse to fine to increase axial resolution of estimated
elastography [9]. Other studies investigated, strain parameters [10], motion decorrelation [11],
refinement process [12], Bilinear Deformable Block Matching (BDBM) technique [13], correlation
parameters were also introduced[14], Integrated the main classification parameters in 3D simulation
for breast tumor [15]. The main aim of this study is to develop new algorithm to estimate elasticity
image that’s uses ultrasound radiofrequency data for breast phantom. The basis of elastic map is
speckles that are used cross correlation method as similarity measurement. The dissimilarity in
displacement map with deep of soft tissue and smallest correlation window length before
compression produced noise in strain image as outliers. Several methods introduced to re-correlated
outliers in order to estimate gradient map.
In this study, the multilevel estimation is introduced to regulate displacement outliers at correlation
window length of 2λ, where refinement operation [16], [17] unable to regularize outliers that
occupies surrounding neighbors particularly at correlation window length of 2λ. However, the
produced algorithm is designed to perform regulation in a conjunction with re-correlation operation
of global stretching and refinement operation, where at first level window length of 3λ is used to
estimate displacement field to replace estimated outliers at second level of window length of 2λ.
The rest of the work is organized as describes the material and methods of the proposed method, then
results and discussion will be demonstrated, finally the conclusion will be highlighted.
2. Material and Methods
Data of before and after applied freehand gentile compression (deformation) were acquired using
ultrasound system model of VINNO-G55 [18]. The compression was applied by transducer physical
surface over the phantom breast model CIRS-59, where the phantom breast manufactured for
elasticity imaging [19]. The experiment of imaging was performed at Al-Mubda’a office in Baghdad-
IRAQ in a conjunction with the research and development department of VINNO Company. In the
experiment, a linear probe model F4-12L broadband of 512 element and 12 MHz was used to acquire
the radiofrequency data [20]. The radiofrequency data were then converted using provided VINNO’s
software tool to be readable at MATLAB workspace for off-line processing as shown in Figure 1.
The acquired radiofrequency data were then processed to estimate displacement and then strain
fields. The proposed algorithm of data processing aims to reduce displacement estimation outliers at
estimation window length of 2λ (20 samples). The proposed algorithm estimates the displacement
field for two sequential level of estimation, first level at window length of 3λ (30 samples) and then
second level at 2λ window length. At the first level, correlation window length and window overlap
ratio are considered based on the investigation study [14]. The first level estimated displacement
outliers are refined based on the previous proposed method of refinement [16] that refine outliers at
the eight neighbor displacement cells as in Figure 2.
In the second level, the estimation displacement outliers are increased dramatically as window length
decreased to 2λ, where refinement method unable to refine some displacement field regions due to
the fact that refinement operation fails when displacement outliers occupies most of eight neighbor
cells [17]. From that end, the proposed method introduced in this study regulates the estimated
displacement outliers occurred at second level of estimation, where the estimated displacement cells
contains outlier replaced by refined displacement of the first level. The refined displacement field of
the first level was already interpolated to match the estimated displacement field dimension of the

The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
3
two levels, as shown in Figure 3. To detect the displacement outliers at second level, metric detection
based on consistency measurement denoted by C is determined, where each cell of displacement
estimation in the second level denoted by S was compared with that one at the first level denoted by
U as in equation 1.
The outliers occurrence at second level of estimation arises when performed cross correlation
estimation of 20 samples window, where the ratio of decorrelation samples is higher than cross
correlation estimation of using 30 samples. Hence, consistency decision is considered based on an
empirical threshold, which is called as a consistency regulation threshold denoted by . When
metric detection C exceeds the consistency threshold , non-consistent estimation declared, and
virus-versa as in equation 2. Consequently, each non-consistent estimation replaces with refined
estimation cell of first level as in Figure 3. Further processing is also considered at second level of
estimation [20] that follows replacement operation, where refinement operation and smoothing
operation are also applied to the regularized displacement field to reduce small scale variation in
displacement estimation, as in the flowchart of the proposed regulation process of Figure 4.
Finally, strain field estimated based on gradient operation as in equation 3, where the strain denoted
by Gs, d1 and d2 represents the two sequentially estimated axial displacements, while l represents the
correlation window length [17]. The parameter of estimation of correlation window length of before
deformation, correlation window after deformation, and window over lapping are 3λ, 1.2, and 0.75
respectively, while at the second stage are 2λ, 1.2, and 0.75 respectively.
  
 (1)
   
   (2)

 (3)
Figure 1. Block daigram of experiment set-up.

The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
4
Figure 2. Identification of the eight neighbors for different CUI locations [17].
Figure 3. Outlier replacement operation, (a) displacement at correlation window length of V=20, (b)
displacement at correlation window length of V=30, (c) displacement at correlation window length of V=30
after interpolation.

The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
5
Figure 4. Flowchart of proposed algorithm.
3. Results and Discussion
The proposed algorithm that has been illustrated in the flowchart shown in Figure 4, considers
multilevel estimation of displacement map as explained earlier, parameters of both correlation and
refinement operation have to be adjusted. In Figure 5(a), displacement map at correlation parameters
represented by, window length of before deformation denoted by was 20 samples, window length
factor of after deformation denoted by F was 1.2, correlation window over lapping denoted by W was
0.75. While the refinement parameters represented by, consistency decision threshold denoted by Tcd
was 4, rate of row detection was 0.03, rate of up and down row detection was 0.06, rate of spatial row
detection was 0.09, and the threshold of reconstruction process denoted by Trec was 0.2. In which, the
displacement field contains outliers located at deep regions that occurs out of estimation range as in
the strain field in Figure 6(a). On the other hand, displacement field in Figure5(b) at =30 samples
shows outlier free, but the strain field in Figure 6(b) shows less axial resolution compared to that one
at =20 samples in Figure 6(a). As a result, when window length increased displacement estimation
occurred as gradient and smooth, which is in turn produces strain field within the range of estimation.
From that end, the smooth displacement field of Figure 5(b) going to be used to regularize the
estimation outliers at =20 samples Figure5 (a), by making detection of estimation outliers and then
regularized them based on first level estimation, as shown in Figure 5(c), but it’s not gradient enough
to produce smooth elastography as shown in Figure 7(a). Hence, to produce smooth elastography at
small window length of 20 samples, refinement operation has to be included within the second level
of estimation. Figure 5(d) shows how the displacement field improved to be gradient, and produces
smoothed elastography as shown in Figure 7(b).

The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
6
Figure 5. Displacement field for correlation window length, (a) refinement of displacement V=20., (b)
refinement of displacement V1=30, (c) multilevel process of displacement V=20.,, (d) multilevel process and
refinement of displacement V=20.
Figure 6. Estimated elastography with Refinement operation at, (a) V1=20, (b) V1=30.

The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
7
Figure 7. Estimated elastography, (a) V1=20 multilevel without refinement operation, (b) V1=20 multilevel
with refinement operation.
4. Conclusion
In this study, radiofrequency ultrasound data are recorded using VINNO G-55 clinic machine to
differentiate hard inclusion that embedded in soft tissue background of phantom breast CIRS-59.
Multilevel estimation algorithm is proposed to regularize displacement estimation outliers at
correlation window length of 20 samples (2λ). The proposed algorithm running in a conjunction with
refinement operation that was introduced before, where refinement process unable to alleviate an
occurrence of outliers occupies surrounding neighbors. Results show the robustness of proposed
algorithm on producing gradient and smooth outlier free displacement field at correlation window
length of 2λ, also shows differential strain field that can recognize hard inclusion at higher spatial
resolution than using correlation window length of 3λ. The findings suggested that using
radiofrequency data for malignant patient’s breast to quantify the robustness of the proposed
algorithm of multilevel estimation.
Acknowledgements
Authors would like to introduce an appreciation to the VINNO Company for given a permission to
access to the radiofrequency of ultrasound clinic machine G-55. Authors are also grateful to Medical
appliance bureau of Alsaaeda for providing the breast model CIRS-59, and to people who
participated in this study.
References
[1] Guo R, Lu G, Qin B and Fei B 2018 “Ultrasound Imaging Technologies for Breast Cancer
Detection and management ” Ultrasound Med. Biol vol 44 no 1 pp 296-301.
[2] Khandpur R S 2003 “Hand Book of Biomedical Instrumentation ” Tata McGraw-Hill Chapter–23
p.623.
[3] Jensen and Jørgen A 1991 “A model for the propagation and scattering of ultrasound in tissue"
Acoustical Society of America Journal vol 89 no 1 pp 182–190.
[4] Bilgen M and Insana M F 1996 “Deformation models and correlation analysis in elastography”
Journal of the Acoustical Society of America vol 99 no 5 pp 3212–3224.
[5] Varghese T, Ophir J and Céspedes I 1996 “Noise reduction in elastograms using temporal
stretching with multicompression averaging” Ultrasound Med. Biol vol 22 no 8 pp 1043–1052.
[6] Chaturvedi P, Insana M F and Hall T J 1998 “2-D companding for noise reduction in strain

The Fourth Postgraduate Engineering Conference
IOP Conf. Series: Materials Science and Engineering 745 (2020) 012105
IOP Publishing
doi:10.1088/1757-899X/745/1/012105
8
imaging” IEEE Trans. Ultrason. Ferroelectr. Freq. Control. vol 45 no 1 pp 179–191.
[7] Insana M F, Chaturvedi P, Hall T J and Bilgen M 2002 “3-D companding using linear arrays for
improved strain imaging” IEEE Ultrasonic Symposium no 97 pp 1438–1435.
[8] Alam S K, Ophir J and Konofagou E E 2002 “An adaptive strain estimator for elastography”
IEEE Trans. Ultrason. Ferroelectr. Freq. Control vol 45 no 2 pp 461–472.
[9] Pellot B C, Frouin F, Insana M F and Herment A 2010 “Ultrasound elastography based on
multiscale estimations of regularized displacement fields” IEEE Transactions on Medical
Imaging vol 23 no 2 pp 153–163.
[10] Hoyt K, Forsberg F and Ophir J 2005 “Investigation Of Paramatric Spectral Estimation
Tecghniquies For Elasticity Imaging” World Federation for Ultrasound in Medicine & Biology
vol 31 no 8 pp 1109–1121.
[11] Liang T, Yung L and Yu W 2013 “On Feature Motion Decorrelation in Ultrasound Speckle
Tracking” IEEE Transactions on Medical Imaging vol 32 no 2 pp 448–435.
[12] Ashikuzzaman M, Gauthier C J and Rivaz H 2019 “Global Ultrasound Elastography in Spatial
and Temporal Domains” IEEE Transactions on Ultrasonic, Ferroelectrics, and Frequency
Control vol 66 no 5 pp 1–12.
[13] Slimi T, Moussa I M, Kraiem T and Mahjoubi H 2018 “New Technique for Processing Tissue
Displacements Estimation in Breast Ultrasound Elastography" Science & Engineering vol 213
no 3 pp 203–233.
[14] Al-jader A R, Al-azawi A, and Ahmed T 2019 “On Ultrasound Elasticity Imaging : Investigation
of Correlation Window Length for Breast Phantom Inclusion Differentiation Using Ultrasound
Radiofrequency Data” The conference on modern electrical and mechanical engineering
systems and applications (MEMSA) pp.7–13.
[15] Naser M A, Sayed A M, Wahba A A and Eldosoky M A 2019 "Modeling Procedures for Breast
Cancer Diagnosis based on Clinical Elastography Images" IEEE International Conference on
Computer Engineering and Systems (ICCES) vol 987 no 1 pp 5110-5386.
[16] Al-azawi A, Dziewierz J, Soraghan J, Inglis S and Gachagan A 2017 "Ultrasound Elastography :
Performance Evaluation of Displacement Field Refinement Process for Freehand Compression”
The International Conference for Engineering Researches (ICER) pp 225–230.
[17] Alazawi A 2015 “On Ultrasound Elastography : Simulation, Experiment, and Algorithm
Development" PhD dessertation University of Strathclyde.
[18] Ashfaq M, Brunke S, Dah J, Ermert H, Hansen C and Insana M 2006 “An ultrasound research
interface for a clinical system” IEEE Trans. Ultrason. Ferroelectr. Freq. Control. vol 53 no 10
pp 1759–1770.
[19] CIRS 2013 “The Perfect Demonstration Tool for Sonoelastography” no 1.
[20] VINNO Transducer 2018 “Specification Ultrasound System Transducers” pp 5–7.