You are on page 1of 6

2020 17th International Multi-Conference on Systems, Signals & Devices (SSD'20)

Evaluation of a novel ultrasound imaging methods


for high frame rate echocardiography
Wided Hechkel Brahim Maaref Néjib Hassen
Laboratory uEI Laboratory EuE Laboratory uEI
University of Monastir University of Monastir University of Monastir
Monastir, Tunisia Monastir, Tunisia Monastir, Tunisia
wided.hechkel@isimm.u-monastir.tn brahim.maaref@isimm.rnu.tn nejib.hassen@fsm.rnu.tn

Abstract—In this paper, we evaluate two new methods of with such a system is the product of the pattern of the
ultrasound imaging that could be used in cardiac imaging. These transmitted beam with the receive pattern of each formed
methods are Parallel Divergent Waves in the same direction and beam.
Parallel Divergent Waves in different directions. We compare
these methods with Consecutive Divergent Waves in different
2020 17th International Multi-Conference on Systems, Signals & Devices (SSD) | 978-1-7281-1080-6/20/$31.00 ©2020 IEEE | DOI: 10.1109/SSD49366.2020.9364257

directions using a compounding option. The goal of our work is


to increase the frame rate of the ultrasound system. Simulation
results prove that image quality for parallel emissions could be
at the same level as consecutive emissions but without estimating
interference, crosstalk and cross-correlation between signals.

Keywords—divergent wave, ultrasound, echocardiography,


frame rate, simultaneous emissions

I. INTRODUCTION
The most demanding requirement of pulse-echo ultrasonic
imaging such as echocardiography can be identified as frame
rate, In the classical sequential-data acquisition scheme, the
frame rate is limited by the imaging depth range and the
number of ultrasonic fires [1][2][3]. A current trend is an
increase in this number to push towards higher imaging
resolution. for example, for a classical 10 cm imaging range,
the relatively slow speed of sound in soft tissue (1540 m/s)
limits pulse-echo ultrasound to a pulse repetition frequency
Fig. 1. The simultaneous synthesis of several transmit/receive events (here
or about 7700 transmit pulses per second in order to avoid 3) requires an independent delay and sum structure for each beam
interferences between successive pulses. In conventional
high-speed B-mode scan imaging, the 7700 transmitted II. METHODS
pulses are typically allocated to about 100 scan lines, giving
rise to a maximal frame rate of 77 frames per second A. Experimental setup
[4][5][6]. Most of the time, this comfortable frame rate is A 32-element, 3.5-MHz 1-D phased-array transducer
reduced at the benefit of resolution: an increased number of transmitting a Gaussian pulse with a bandwidth of 50% was
imaging lines and several ultrasonic fires per line (i.e. several simulated in this study. The transducer measured 5 mm in
transmit foci) [7][8][9]. Several techniques have been height and 7.7 mm in width, with a pitch of 0.242 mm and
proposed to increase the imaging rate. In the first category of with a kerf of 0.022 mm. The two-way beam profiles of the
techniques, the sequential acquisition scheme is retained: the following systems were simulated: 1) 64MLA using several
pulse repetition frequency is optimized through the parallels transmit diverging-wave imaging with a virtual
minimization of the imaging range with the use of focal depth of 100 mm with different directions 2) 64MLA
endoprobes. These techniques are rather interesting in the using several parallels transmit diverging-wave imaging with
sense that they give rise to very high resolution and high a virtual focal depth of 100 mm at the same direction ;3)
frame rate at the same time. however, they concern a small 64MLA using several parallels transmit diverging-wave
number of organs (e.g. heart, Prostata. ...) and they are less imaging with a virtual focal depth of 100 mm with different
flexible than peripheral exams. In the second category of' directions with spatial compounding. The implementation
techniques, a parallel acquisition scheme is introduced details of these beamforming strategies are given in Table I.
[10][11]. Its principle is rather simple and was recognized For the beam profile at a particular steering angle, the field of
early in phased array imaging. A rather wide illumination view was set to different azimuth directions relatively to each
beam is transmitted using multiple pulse generators (figure divergent beam (Figure 2). For all the imaging systems
1.a). Upon reception, several beams can be synthesized investigated, dynamic focusing was employed to receive
simultaneously, using the same array data, by applying a using the full aperture and Hamming apodization was used to
different time delay law for each independent beam. As it is receive to suppress the side lobes.
apparent in figure 1.b, this technique requires rather complex
hardware to simultaneously handle the formation of several
beams. The pulse-echo beam pattern of each beam produced

978-1-7281-1080-6/20/$31.00 ©2020 IEEE 185

Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on May 16,2021 at 17:07:26 UTC from IEEE Xplore. Restrictions apply.
 Signal to Noise Ratio: The SNR as a function of
depth, z, can be determined as the ratio of the signal
TABLE I. Configuration parameters of the simulations power, S, to the noise power, N [13]:
Parameter Symbol Value
SNR (z) = = (1)
-
Speed of sound in tissue c 1540 m/s
where E is the expectation operator, and x is the
Transducer center f0 4 MHz
frequency signal radiofrequency (RF) value.
Transducer bandwidth B 4 MHz In this article, the signal power will be estimated
Wavelength λ 0.385mm from M measurements as:
Sampling frequency fs 20 MHz S(z) = ∑ (2)
Transducer array size N 32 and the noise power will be estimated as: N(z) =
(2D) elements
Element directivity in 0.707 rad ∑ ∑ (3)
the1D Array probe
(acceptance angle) SNR expressed in dB is calculated in accordance
with
SNRdB(z) = 10 * log10(SNR(z)) (4)
 Contrast Per Pixel (CPP): CPP is an estimation of
the average intensity difference between a pixel and
its adjacent pixel. The CPP of an image is defined to
be [14]:
∑ ∑ ∑ , | , , |
,
(5)

Here, f(i,j) is the gray value of pixel (i,j), and f(m,n)


is the gray value of neighboring pixel (i, j) in the
Fig. 2. Different azimuth directions relatively to each divergent beam
image.
B. Software Ultrasound Simulator  Contrast to Noise Ratio (CNR): CNR is defined as
Numerical simulations (in Field II simulation Program) of the logarithmic difference between the mean values
ultrasound imaging were performed first in order to test
various configurations and appropriately choose the best ones of the image kernels fully within and outside the
for the final ultrasound probe design. The possible cysts and is expressed as [15]:
configurations of the ultrasound probe were tested and the
image quality analyzed. The media whose images have been CNR = (6)
simulated included is Point Spread Function (PSF).
where μ denotes the mean and σ denotes the
C. Image Quality Metrics for PSF and PSFs standard deviation of the log-compressed B-mode
High image quality is a critical purpose for the building image. The subscripts T and B stand for the target
of a good ultrasound imager. Consequently, accurate analysis
depends heavily on the reconstructed output image quality. and the background, respectively.
The most important metric to evaluate the spatial resolution
is Full Width at Half Maximum (FWHM). In this work, we III. EXPERIMENTAL EVALUATION
use a new evaluation metric [12], Full Width at Half Dynamic
Range (FWHDR). The motivation of choosing a metric A. Point Spread Functions Results
related to dynamic range (DR) is the fact that the image As you can see from figure 3(b), we used ten PSFs in our
resolution drops in high DR levels and it becomes difficult to synthetic phantom, simulation results for one divergent wave
differentiate between closely placed scatterers. are seen in figure 3(a). Figure 4 shows simulation results for
the three algorithms, PDW-SD, PDW-DD, and CDW-DD-C.
D. Cyst Phantom Evaluation Metrics Results are reobtained for consecutively three, five, seven,
Metrics used to evaluate the quality of the Cyst Phantoms nine, eleven and thirteen divergent waves.
(CP) are: Table 2 shows the Full Width at Half Maximum (FWHM) in
millimeter (mm) for lateral resolution for each algorithm.

186

Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on May 16,2021 at 17:07:26 UTC from IEEE Xplore. Restrictions apply.
Here we observe that PDW-SD shows the same value for all hence, for every transmit/receive (tr/rc) event, (10*10-2/1540)
simulations (0.0217 mm). Here, all insonifications are at the second is taken, so, 6.5 is taken. Hence, if we construct
same direction, so, the FWHM cannot well evaluate the one image every one tr/rc event, the frame rate was 7700 Hz
spatial resolution of the image, because, it gives at all (frame/second). So, by the same logic, we obtained the results
simulations the same values, because, the implementation of table 5, were div 1 has 7700 Hz for all techniques, also
logic of the algorithms obliges us to sum the response of each parallel transmissions have the same value (7700 Hz) and
transmitted wave and to superimpose the results. finally, for consecutive transmissions, the frame rate decrease
Also, we observe from the same table that PDW-DD and with the increase of the number of transmitted divergent
CDW-DD-C have the same values for each simulation, this waves.
is due because the compounding option used in CDW-DD-C
is the average between all the received responses of each
insonification, and parallel transmissions in PDW-DD were
treated as consecutive transmissions, the case with CDW-
DD-C because of simulator constraints. The best values here
are highlighted in the table, div 3 with 0.0145 mm, div 5 with
0.0238 mm and div 9 with 0.0558 mm.
Table 3 shows the Full Width at Half Dynamic Range
(FWHDR) in millimeter (mm) for lateral resolution for each
algorithm. We observe also the same remarques as FWHM.
PDW-SD has 0.1695 mm, best values for PDW-DD AND
CDW-DD-C are 0.1462 mm for 3 did; 0.1567 mm for 5 div
and 0.1681 for 7 div.
Table 4 shows the execution time for each algorithm.
Certainly, the minimum values are obtained with algorithms Fig. 3. (a) Simulation result for one divergent Beam (b) Synthetic phantom
that have a smaller number of insonifications. So, the final containing ten PSFs
image for 1 div algorithm is taken after about 35 seconds; div
3 for PDW-SD, PDW-DD, and CDW-DD-C are taken
consecutively after 71, 103 and 70 seconds.
Table 5 shows the frame rate for each algorithm, here the
frame rate was obtained standing to the equation above,

Fig. 4. Point Spread Functions Simulations Results for consecutively (a) three (b) five (c) seven (d) nine (e) eleven (f) thirteen divergent waves, if option 1:
CDW-DD-C else if 2 PDW-DD else if 3 PDW-SD.

187

Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on May 16,2021 at 17:07:26 UTC from IEEE Xplore. Restrictions apply.
TABLE II. FULL-WIDTH HALF MAXIMUM (LATERAL RESOLUTION) OF THE SIMULATIONS (MM)

Algorithm PDW-SD PDW-DD CDW-DD-C


DIV 1 0.0217
DIV 3 0.0217 0.0145 0.0145
DIV 5 0.0217 0.0238 0.0238
DIV 7 0.0217 0.0573 0.0573
DIV 9 0.0217 0.0558 0.0558
DIV 11 0.0217 0.0639 0.0639
DIV 13 0.0217 0.0634 0.0634

TABLE III. FULL-WIDTH HALF DYNAMIC RANGE (LATERAL RESOLUTION) OF THE SIMULATIONS (MM)

Algorithm PDW-SD PDW-DD CDW-DD-C


DIV 1 0.1695
DIV 3 0.1695 0.1462 0.1462
DIV 5 0.1695 0.1567 0.1567
DIV 7 0.1695 0.1681 0.1681
DIV 9 0.1695 0.1705 0.1705
DIV 11 0.1695 0.1723 0.1723
DIV 13 0.1695 0.1718 0.1718

TABLE IV. EXECUTION TIME OF THE SIMULATIONS (SECOND)

Algorithm PDW-SD PDW-DD CDW-DD-C


DIV 1 35.023
DIV 3 71.585 103.636 70.703
DIV 5 112.892 131.488 108.204
DIV 7 145.771 168.676 144.430
DIV 9 184.806 207.785 190.414
DIV 11 230.118 245.752 218.986
DIV 13 272.175 279.245 262.512

TABLE V. FRAME RATE OF THE SIMULATIONS (HZ)

Algorithm PDW-SD PDW-DD CDW-DD-C


DIV 1 7700

DIV 3 2567

DIV 5 1540

DIV 7 1100
7700
DIV 9 856

DIV 11 700

DIV 13 592

188

Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on May 16,2021 at 17:07:26 UTC from IEEE Xplore. Restrictions apply.
B. Cyst Phantom Results 11 div. Further, for PDW-SD, the best SNR value is 21.15 dB
Figure 5 shows the B-mode cyst phantom images for for 11 div also.
CDW-DD-C, PDW-DD and PDW-SD techniques. The By the same sense, PSNR best values are -35.80 dB, -28.09
results are obtained consecutively for three, five, seven, nine, dB and -28.28 dB for consecutively CDW-DD-C / 5div
eleven and thirteen divergent waves. Table 6 demonstrates PDW-DD / 9div and PDW-SD / 3div.
further performance evaluation for each algorithm with the The logic for CNR and CPP calculations is to evaluate all
calculation of the parameters that describe the image options. the images by the same level without using a reference image.
For SNR and PSNR we take 13 div of each algorithm as a So, the three best values are highlighted in table 6. For CNR,
reference to the other simulations of the same algorithm. they are 30.68 dB, 32.43 dB and 32.54 dB for consecutively
Hence, for CDW-DD-C, the best SNR value is 9.54 dB for 5 PDW-SD / 13div PDW-SD / 11div and PDW-SD / 5div. For
div. Also, for PDW-DD, the best SNR value is 23.41 dB for CPP, they are 19.98, 19.96 and 19.97 for consecutively PDW-
DD / 9div, PDW-SD / 11div and PDW-SD / 5div.

Fig. 5. Cyst Phantom B-mode Simulations Results for consecutively (a) three (b) five (c) seven (d) nine (e) eleven (f) thirteen divergent waves, if option 1:
CDW-DD-C else if 2 PDW-DD else if 3 PDW-SD.

TABLE VI. PERFORMANCE EVALUATION METRICS FOR CYST PHANTOM B-MODE SIMULATIONS

Performance metrics of Cyst Phantom


Algorithms
SNR (dB) PSNR (dB) CNR CPP
(dB)
CDW-DD-C / 13div Reference Reference 29.26 17.78
CDW-DD-C / 11div 9.17 -35.48 29.34 19.91
CDW-DD-C / 9div 9.39 -35.25 29.00 19.73
CDW-DD-C / 7div 9.15 -35.50 29.41 19.94
CDW-DD-C / 5div 9.54 -35.80 28.26 17.58
CDW-DD-C / 3div 9.16 -35.49 29.09 19.95
PDW-DD / 13div Reference Reference 29.69 19.77
PDW-DD / 11div 23.41 -21.89 29.96 19.74
PDW-DD / 9div 17.21 -28.09 28.37 19.98
PDW-DD / 7div 17.39 -27.91 28.31 19.93
PDW-DD / 5div 17.36 -27.94 29.77 19.95
PDW-DD / 3div 17.24 -28.06 29.33 19.86
PDW-SD / 13div Reference Reference 30.68 19.84
PDW-SD / 11div 21.15 -24.19 32.43 19.96
PDW-SD / 9div 19.71 -24.94 24.63 17.44
PDW-SD / 7div 21.56 -23.78 30.42 19.78

189

Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on May 16,2021 at 17:07:26 UTC from IEEE Xplore. Restrictions apply.
PDW-SD / 5div 20.91 -24.43 32.54 19.97
PDW-SD / 3div 17.06 -28.28 29.85 19.59

[2] M. Couade, M. Pernot, M. Tanter, E. Messas, A. Bel, M. Ba, A.-A.


IV. DISCUSSION Hagege, and M. Fink, “Ultrafast imaging of the heart using circular
wave synthetic imaging with phased arrays,” in IEEE International
Simulation results prove that FWHM and FWHDR of the Ultrasonics Symposium (IUS), IEEE International, 2009, pp. 515–518.
parallel divergent waves in the same direction algorithm are [3] L. Tong, H. Gao, H. F. Choi, and J. D’hooge, “Comparison of
the same as only one divergent wave, and this is caused by conventional parallel beamforming with plane wave and diverging
the superposition of the beamformed rf_data without the wave imaging for cardiac applications: a simulation study,” IEEE
Trans. Ultrason. Ferroelectr. Freq. Control, vol. 59, no. 8, pp. 1654–
estimation of the interference in the image. FWHM and 1663, Aug. 2012.
FWHDR of the parallel divergent waves at different direction [4] L. Tong, H. Gao, and J. D’hooge, “Multi-transmit beam forming for
algorithms and the consecutive divergent waves at different fast cardiac imaging-a simulation study,” IEEE Trans. Ultrason.
directions with compounding algorithms are also the same Ferroelectr. Freq. Control, vol. 60, no. 8, pp. 1719– 1731, Aug. 2013.
and this is because the compound option that we used is the [5] L. Petrusca, F. Varray, R. Souchon, A. Bernard, J.-Y. Chapelon, H.
Liebgott, W. A. NDjin, and M. Viallon, “Fast volumetric ultrasound b-
average between the different received beamformed echoes. mode and doppler imaging with a new high-channels density platform
Also, in this case, we cannot say that these results perfectly for advanced 4d cardiac imaging/therapy,” Applied Sciences, vol. 8,
estimate the difference between algorithm because the no. 2, 2018.
simulator cannot estimate crosstalk and cross-correlation [6] M. Correia, J. Provost, S. Chatelin, O. Villemain, M. Tanter, and M.
between the simultaneous firings in the case of parallel waves Pernot, “Ultrafast harmonic coherent compound (uhcc) imaging for
high frame rate echocardiography and shear-wave elastography,” IEEE
at different directions. For the execution time, we clearly Transactions on Ultrasonics, Ferroelectrics, and Frequency Control,
remark that it increases with the increase of the number of vol. 63, no. 3, pp. 420–431, 2016.M. Young, The Technical Writer’s
waves. In the case of the frame rate, we can easily prove that Handbook. Mill Valley, CA: University Science, 1989.
one divergent wave and simultaneous divergent waves have [7] E. Bădescu, “High-frame rate ultrasound methodologies for cardiac
applications”. Medical Imaging. Université de Lyon, 2018. Thesis
the same frame rate, but the frame rate decrease if the number 144p.
of the waves increases in case of consecutive insonifications. [8] H. Hasegawa, H. Kanai, “High-frame-rate echocardiography using
Cyst phantom simulation proves that PDW-SD /11 div has at diverging transmit beams and parallel receive beamforming” J. Med.
many times the best performance metrics, so it could be the Ultrason. Jul.2011 38:129–140.
best algorithm for later improvements and implementation. [9] R. Mallart and M. Fink, “Improved imaging rate through simultaneous
transmission of several ultrasound beams,” 1992, pp. 120–130.
V. CONCLUSION [10] E. Tiran, T. Deffieux, M. Correia, D. Maresca, B.-F. Osmanski, L.-A.
Sieu, A. Bergel, I. Cohen, M. Pernot, and M. Tanter, “Multiplane wave
In this paper, we simulated two new ultrasound algorithms imaging increases signal-to-noise ratio in ultrafast ultrasound
for B-mode phased array echocardiography in order to imaging,” Physics in Medicine and Biology, vol. 60, no. 21, pp. 8549–
evaluate the performance of these algorithms in comparison 8566, 2015.
with conventional imaging algorithm that uses consecutive [11] M. Tanter and M. Fink, "Ultrafast imaging in biomedical ultrasound",
diverging wave and compounding. Image quality metrics help IEEE transactions on ultrasonics, ferroelectrics, and frequency control.
us to define the quality of the image for each algorithm and Vol. 61(1), pp. 102-119. 2014.
execution time is the best index that qualifies the [12] S. Harput, J. McLaughlan, D. M. Cowell, and S. Freear, “New
performance metrics for ultrasound pulse compression systems,” in
computational complexity of the algorithms. Also, the frame IEEE International Ultrasonics Symposium (IUS), 2014, pp. 440–443.
rate for the different techniques helps us to more distinguish [13] V. Papic, Z. Djurovic, G. Kvascev, P. Tadic, “On signal-to-noise ratio
the reliability of each of them. But simulation results don’t estimation.” In Proceedings of the Malecon 2010—2010 15th IEEE
estimate many options in the image like interference, cross Mediterranean Electrotechnical Conference IEEE, Valletta, Malta,25–
talk, and cross-correlation. Thus, in the next steps, we will 28 April 2010, Volume 5, pp. 160–165.
implement these approaches in our digital beamforming [14] M. Eramian, D.Mould, ” Histogram equalization using neighborhood
ultrasound system and evaluate them in a real system. metric.” Proceedings of the Second Canadian Conference on Computer
and Robot Vision; 2005 May 9–11; Victoria, BC, Canada. Washington:
REFERENCES IEEE Computer Society; 2005.
[15] M. Welvaert, Y. Rosseel, “On the definition of signal-to-noise ratio and
[1] J. Bercoff, “Ultrafast ultrasound imaging,” in Ultrasound Imaging, contrast-to-noise ratio for fMRI data.” PLoS One. 2013, 8: e77089.
Medical Applications, O. Minin, Ed. Rijeka, Croatia: InTech, 2011, pp.
3–24.

190

Authorized licensed use limited to: Univ of Calif Santa Barbara. Downloaded on May 16,2021 at 17:07:26 UTC from IEEE Xplore. Restrictions apply.

You might also like