Content uploaded by Luc Jonveaux
Author content
All content in this area was uploaded by Luc Jonveaux on Jan 09, 2020
Content may be subject to copyright.
48
Development of a Portable and Inexpensive Ultrasound
Imaging Device for Use in the Developing World
Zach Taylor, Luc Jonveaux, Charles Caskey
Vanderbilt University Institute of Imaging Science, Department of Radiology and Radiological Sciences, Vanderbilt University, Nashville, TN
KEYWORDS. BeagleBone, Phantom, Portable, Ultrasound.
BRIEF. A portable and inexpensive ultrasound device that could be used in the developing world was developed and tested.
ABSTCT. People in developing countries have limited access to life-
saving diagnostic equipment. Because medical imaging devices are sta-
tionary and costly, there exists a need for imaging technology that is not
only accurate and portable, but also inexpensive. To address this issue, we
developed and tested an inexpensive portable ultrasound device. ree
microprocessing boards compose the device: a SeeedStudio BeagleBone
Green, an Arduino Uno, and a Murgen board. e BeagleBone powers
and controls the Murgen board. e Murgen board pulses a 5MHz single-
element transducer, rotated by the Arduino, and receives the echoes. We
programmed acquisition and image reconstruction procedures for the
device and assessed the signal-to-noise ratio (SNR) in images of high-con-
trast graphite laboratory phantoms as well as standard clinical phantoms
manufactured by Computerized Imaging Reference Systems, Inc. (CIRS).
Reconstructed images of laboratory phantoms yielded an SNR of 9.3 dB,
which was acceptable for imaging high-contrast targets. Some targets in the
CIRS phantom were visible, but the SNR remained below an acceptable
threshold, revealing the need for additional signal processing and noise
reduction. All in all, we have demonstrated the feasibility of, identied fur-
ther improvement for, and laid the foundation for an inexpensive portable
ultra-sound device.
INTRODUCTION.
Accurate medical imaging is necessary for proper diagnoses as patients oen do
not show outward symptoms until it is too late for treatment. However, people
in developing countries have limited access to life-saving diagnostic methods,
oen having to travel long distances to larger, more auent cities for medical
care [1]. A mobile and low-cost imaging option has the potential to benet
patients in remote areas who have limited access to such devices. Because ultra-
sound does not require large scanners like magnetic resonance imaging (MRI)
or computerized tomography (CT), it is the ideal modality for a portable imag-
ing device. In this project, an ultrasound device that is inexpensive, portable,
and accurate has been developed and tested.
Ultrasound devices transmit and receive sound waves via a transducer. e time
dierence between wave transmission and reception can be mapped to a one-
dimensional image. is type of imaging is called one-dimensional (A-mode)
imaging. ough A-mode imaging is inexpensive to implement because it uses
only one transducer, it has limited use for medical diagnostics. Two-dimensional
(B-mode) images are more widely used in medical ultrasound applications,
such as prenatal, trauma, and cancer imaging [2]. However, B-mode devices
usually consist of multiple transducers, each with its own receive circuit, and
thus are prohibitively expensive. In order to achieve two-dimensional imag-
ing while maintaining a low cost, in our device the transducer is rotated, and
one-dimensional images are taken in rapid succession as the transducer sweeps
through a 60-degree angle. is is called sector scanning.
Sector scanning can be used to generate B-mode images while helping keep the
device low-cost. Furthermore, integrated circuits exist for all necessary com-
ponents of an ultrasound scanner, including transmit-receive switches, noise
ampliers, and ADC converters. ese can be combined with minimal com-
puting power, with the potential to cost less than $300. In this study, a single-
element ultrasound device was developed and tested in order to assess the fea-
sibility of generating accurate ultrasound images for a low cost.
Portable ultrasound has been the focus of various recent research studies [1],
one of which was tested in a Level-I trauma hospital in Detroit, Michigan.
Kirkpatrick et al. developed and tested the eectiveness of a portable ultra-
sound device (HHFAST) to perform Focused Assessment with Sonography
for Trauma (FAST) ultrasound exams [3]. e HHFAST Sonosite 180, a 2.4kg
portable ultrasound scanner, was implemented in a hospital triage seing with
a 97% accuracy in predicting the clinical outcome. Kirkpatrick et al. shows that
a small and portable ultrasound device would be feasible for geing clinically
useful and valuable data that could be used for diagnosis. While the Sonosite
imaging device provides an accurate and portable option, it costs up to $25,000
[4]. e next step towards ubiquitous medical imaging technology is the devel-
opment of inexpensive diagnostic devices.
MATERIALS AND METHODS.
Board Descriptions.
e device consisted of a 5MHz single-element transducer and three open-
source microprocessing boards each with a dierent task to perform: a
SeeedStudio BeagleBone Green, a Murgen board, and an Arduino Uno. e
BeagleBone is a low-cost development platform that runs Debian Linux [5]. It
is powered and controlled the timing of the Murgen board, developed by Luc
Jonveaux et al. as a novel platform for a low-cost ultrasound machine [6] that
functions as transmit-receive switch and high voltage pulser. e Murgen board
also contains circuitry for envelope detection and analog-to-digital conversion,
but testing these components was beyond the scope of this project. e Murgen
board and BeagleBone have high sampling rates and fast image processing. e
Arduino is an open-source microprocessing board. As its limited clock speed of
16MHz [7] is less than the 25MHz sampling rate needed for ultrasound, it was
only used to rotate the Servo motor.
Table 1. Costs and functions of device components.
Component Cost Function
SeeedStudio BeagleBone
Green
$40 Control Murgen Board
Murgen Board $500 Time pulses; process echoes
Arduino Uno $25 Rotate transducer
Transducer $200 Send pulses; receive echoes
Total $765 Collect signal for image reconstruction
Data Acquisition.
e three microprocessing boards are connected to make the whole device
(Fig. 1). e BeagleBone initialized two consecutive pulse width modulation
(PWM) waves that were then sent to trigger the Murgen board at an interval
of 250 microseconds. Upon receiving these trigger waves, the Murgen board
pulsed the transducer, and it then received the resulting pulse-echoes back from
the transducer.
To generate two-dimensional images, the Arduino rotates the transducer via
a Servo motor (Batan model S1213). At each angle in a sweep of 60 degrees,
MATLAB gathers 20 datasets from the Murgen board using GageScope and
averages the datasets to reduce noise.
49
Image Processing.
Datasets were gathered and averaged twenty times in MATLAB using
GageScope with CompuScope, and then the average dataset for each angle
underwent further signal processing and image reconstruction in Python 3.5.
e signal at each angle underwent noise reduction via a Buerworth band-
pass lter and envelope detection via a Hilbert transformation. Since the trans-
ducer sends 5MHz pulses, any received signal that is not between 4.5MHz and
5.5MHz is considered noise and reduced by the Buerworth lter. Envelope
detection ensures that the nal image accounts for the entire body of a solid
object (rather than just its edges) and results in an overall smoother image.
Envelope detection also removes the carrier frequency from the signal in order
to emphasize the locations of the echoes in the nal image. Next, because the
data was collected by rotating a Servo motor, it is converted from polar coor-
dinates to Cartesian coordinates. is was accomplished by mapping a grid of
squares each with a side length of 0.25mm. For each of these square sections in
rectangular coordinates, the corresponding polar coordinates were calculated.
e square section was then assigned the value of the dataset at the calculated
polar coordinates. Finally, the signal is log-compressed so as to further dier-
entiate the envelopes from background noise and therefore boost the image
contrast.
Phantom Imaging and Image Analysis.
Two phantoms were imaged in this study by submerging the transducer in
water. A high-contrast graphite phantom made in-house by a graduate student
and a standard medical phantom manufactured by Computerized Imaging
Reference Systems, Inc (CIRS). e in-house phantom was made by adding
4g graphite and 3g agar to 8mL n-propanol and 92mL water and heating the
solution. e resulting substance is then set in a cylindrical mold and allowed to
set in a refrigerator. is phantom was placed 40mm away from the submerged
transducer face in the tank (Fig. 2a). Similarly, the CIRS phantom was placed
15mm away from the transducer face. To quantify the results, the processed
image was analyzed for SNR in ImageJ between the phantom and the surround-
a)
b)
Figure 2. e ultrasound image of a graphite phantom. e front edge of the phan-
tom is 40mm away from the transducer. Fig. 2a is the setup of a submerged phantom
and 2b is the reconstructed image.
Figure 1. Device schematic. e BeagleBone triggers the Murgen board at a set interval. e Murgen board pulses and receives echoes from the transducer. e Arduino
controls the rotation of the transducer. MATLAB 2013a is used to gather data and control the Arduino.
50
acquisition; using the onboard ADC would thus make the device more por-
table. e BeagleBone would still control the Servo via a serial connection with
the Arduino, which has already been accomplished. Furthermore, the need
for a computer could potentially be eliminated by aaching a screen to the
BeagleBone to show the reconstructed image. e resulting design is common
among portable ultrasound devices, such as the GE VScan [4].
e device currently executes a full sweep and reconstructs an image in about
two minutes. e majority of this time is due to heavy averaging implemented
at each angle during data collection. e incorporation of signal processing on
the front end (i.e., by the Murgen board) would reduce the need for averaging
and thus increase the frame rate. e ultimate goal would be to reduce signal
processing for each angle to less than 10ms, as this is the fastest rate that the
Servo motor can rotate to each angle. e scan conversion is the most time-
consuming function of the image reconstruction program. However, since the
scan conversion could run in parallel with image acquisition, it will not aect
the acquisition time at each angle.
ing water by dividing the mean intensity of the signal by the mean intensity of
the background noise. SNR is then converted to an intensity value in decibels
(Equation 1).
Intensity(dB) = 10 * log10(SNR) (1)
RESULTS.
Image Reconstruction.
2D images were successfully acquired by reconstructing echoes received from
the Murgen board. In the reconstructed image of the graphite phantom (Fig.
2b), the front edge of the phantom is at 40mm. As expected, the signal intensity
from the phantom fades with increasing distance from the transducer. ere
is an SNR of 5.42, with a signal intensity of 7.34 dB. Moreover, it meets Rose’s
criterion, which states that for an image to have 100% clarity, its SNR must be
greater than 5 [10].
Comparing to a Clinical Standard.
e CIRS ultrasound phantom (Fig. 3a) is used as a standard to compare
ultrasound image quality [11]. e Murgen device (Fig. 3b) was compared to
the Verasonics commercial-grade ultrasound machine (Fig. 3c). Both devices
imaged the CIRS phantom, and corresponding targets are circled in red. e
Murgen device yielded an SNR of 2.36, with an intensity of 3.73 dB. Verasonics
had an SNR of 6.29, with an intensity of 7.99 dB. e Murgen device did not
meet Rose’s criterion when imaging the CIRS phantom, though the Verasonics
did. e Murgen device had visible point scaering (Fig. 3b). Moreover, echoes
o the front edge of the CIRS phantom reected, resulting in periodic, dimin-
ishing white bands to appear in the image where no object existed (Fig. 3b).
DISCUSSION.
Imaging the Graphite Laboratory Phantom.
In imaging the graphite phantom, the front face of the phantom was placed
40mm away from the transducer (Fig. 2a). In the re-constructed image (Fig.
2b), the front of the phantom also appears at 40mm, so the distance scale is
accurate. e phantom also appears slightly elongated in the x-direction
because the sweep of the Servo motor was not quite parallel. e phantom
also appears to get darker with distance from the transducer. is phenome-
non called aenuation is common in ultrasound imaging and is corrected for
by incorporating time-gain compensation in the image processing. Time-gain
compensation articially boosts the signal strength with respect to distance
from the transducer [2].
e image of the graphite phantom had a SNR of 5.42, which meets Rose’s cr iteri-
on. erefore, our device can image graphite phantoms with 100% clarity. Further
testing will need to be done to determine its feasibility in imaging tissue.
Imaging the CIRS Phantom.
e Murgen device successfully imaged a commercially standard phantom
(Fig. 3a-c). ere is visible point scaering in the Murgen device. Incorporating
deconvolution in the image processing will correct for this fuzziness.
Deconvolution is used to decrease the fuzziness of the image. e retreating
white bands in the Murgen’s image of the CIRS phantom (Fig. 3b) are reec-
tions of the echo o the phantom’s front edge, a result of the particular arrange-
ment of the transducer and phantom. Furthermore, there was an SNR of 2.36,
not meeting Rose’s criterion. erefore, our device does not image commercial
phantoms with 100% clarity. Targets in the CIRS phantom were expected to
be more dicult to image than the submerged graphite phantoms because it is
used for quality testing.
Device Optimization.
By implementing front-end signal processing on the Murgen board, the nec-
essary computing power of the device reconstructing the images would be
reduced. In the future, data could be gathered directly onto the BeagleBone
using an ADC connection between it and the Murgen. is would eliminate
the necessity of MATLAB/GageScope, which are currently used for data
Figure 3. Comparing images of the CIRS phantom taken by the Murgen and
Verasonics. Fig 3a. is a diagram of the targets in the CIRS Model 040GSE Multi-
Purpose, Multi-Tissue Ultrasound Phantom is a standardization tool for ultrasound
imaging devices. Corresponding targets are circled in red. (Fig. 3a source: [11]).
3b and 3c show the reconstructed image from the Murgen and Verasonics imaging
devices, respectively. Corresponding targets are circled in red.
a)
b) c)
51
In this project, a portable and inexpensive ultrasound device was developed.
e device takes accurate images of laboratory phantoms, but needs further
improvement before it could be commercially implemented. e device costs
97% less than commercial scanners (Table 1), which can cost up to $25,000
[4]. To further drive down cost, a transducer built in-lab could be implement-
ed. is would drive down the cost of the device even further. is project lays
the groundwork for a portable ultrasound device of comparable quality to com-
mercial grade scanners and at a substantially lower price.
ACKNOWLEDGEMENTS.
I would like to thank Tony Phipps for use of his phantoms and for advice through-
out the project, Vandiver Chaplin for help and advice in writing code, Jiro Kusunose
for use of the Verasonics scanner, the Caskey Lab at Vanderbilt University Institute
for Imaging Science, and Bre Byram for use of the CIRS phantom.
REFERENCES.
1. S. Sippel, K. Muruganandan, A. Levine, S. Shah, Review Article: Use of
Ultrasound in the Developing World. International Journal of Emergency Medicine. 4,
111 (2011).
2. A. Kirkpatrick, R . Simmons, R. Brown, S. Nicolaou, S. Dulchavsky, e hand-
held FAST: Experience with hand-held trauma sonography in a level-I urban
trauma center. International Journal of the Care of the Injured. 33, 303-308 (2002).
3. P. Suetens, Fundamentals of Medical Imaging (Cambridge University Press, New
York, ed. 2, 2013), pp. 128-158. [Second edition]
4. “Ultrasound Comparison Guide,” Providian Medical Equipment [Online].
Available: hp://www.providianmedical.com/wp-content/uploads/2016/01/
Providian-Medical-Ultrasound-Machine-Comparisons.pdf [ June 30, 2016].
5. D. Molloy, Ex ploring BeagleBone, Indianapolis, IN: John Wiley & Sons, Inc.,
2015 [Online]. Available: hps://dl.dropboxusercontent.com/u/25361/explor-
ing-beaglebone.pdf [May 26, 2016].
6. L. Jonveaux, et al., “Murgen: An Open-Source Ultrasound Imaging dev-kit side
project,” [Online]. Available: hps://hackaday.io/project/9281-murgen [May 25,
2016].
7. “Arduino Uno,” [Online]. Available: hps://www.arduino.cc/en/Main/
ArduinoBoardUno [June 30, 2016].
8. SeeedStudio BeagleBone Green, [Online]. Available: hp://www.seeedstudio.
com/wiki/images/thumb/1/17/450px-BBG3.jpg/450px-450px-BBG3.jpg
9. Arduino Uno and Servo Motor Fritzing Diagram, [Online]. Available: hp://
dm.ncl.ac.uk/clarerobertson/les/2013/11/sweep_BB.png
10. J.T. Bushberg, J.A. Seibert, E.M. Leidholdt, J.M. Boone, e Essential Physics
of Medical Imaging (Wolters Kluwer Health/Lippinco Williams & Wilkins,
Philadelphia, ed. 3, 2012), pp. 280-281. [ird edition]
11. CIRS, Norfolk, VA [Online]. Available: hp://www.cirsinc.com/le/
Products/040GSE/040GSE%20DS%20101915.pdf
Zach Taylor is a student at Hume-Fogg Academic
Magnet High School in Nashville, TN; he participat-
ed in the School for Science and Math at Vanderbilt.
