VOL. X, NO.

X, XXXXXXXX ISSN 1819-6608

ARPN Journal of Engineering and Applied Sciences
©2006-2013 Asian Research Publishing Network (ARPN). All rights reserved.

www.arpnjournals.com

COMPARATIVE STUDY OF EDGE DETECTION ALGORITHM: VESSEL

WALL ELASTICITY MEASUREMENT FOR DEEP VEIN THROMBOSIS
DIAGNOSIS
Noor Hafizzatul ‘Izzah bt Mat Harun, Nabilah bt Ibrahim, Nur Shazilah bt Aziz
Microelectronics and Nanotechnology – Shamsuddin Research Centre (MiNT-SRC), Electrical and Electronic
Engineering Faculty, Universiti Tun Hussein Onn Malaysia, 86400, Batu Pahat, Malaysia
E-Mail: [email protected], [email protected], [email protected].

ABSTRACT
In this paper, a comparative study of different edge detection methods had been done to determine the vessel wall elasticity for
early diagnosis of the Deep Vein Thrombosis condition. Currently, in most research found that the measurement of the vessel
detection conducted solely on the raw image obtains from the ultrasound. Thus, the precision of the measurement could be
controvertible from time to time. As a matter of fact, the image consists of its individual characteristics or properties that
cannot be verified distinctly. Therefore, various methods of edge detection techniques had been applied to the B-mode
ultrasound image. There are several edge detection techniques available for pre-processing in computer vision. Though,
Canny, Sobel and Roberts are some of the most applied methods. This paper compares each of the methods by the evaluation
of the Mean Squared Error (MSE) and Peak Signal to Noise Ratio (PSNR) of the output image. The evaluations are using
Matlab software, the algorithms applied on the B-mode image of 10 subjects that had been volunteering for the purpose of
the study. Both the MSE and the PSNR were in numeric values, that includes the vessel wall elasticity measurement of the
popliteal vein, therefore, the performance of the algorithm is determined.

Keywords: Edge detection, DVT, MSE, PSNR, vessel wall elasticity.

INTRODUCTION Doppler is useful in diagnosing DVT at risk patient and

Ultrasound imaging had been an important role in provide a noninvasive method of investigation. In addition,
diagnosing various kinds of health condition. This has it is also contributes in evaluating the site, extent and stage
included for the diagnosing of the Deep Vein Thrombosis of thrombus. As proven, ultrasound imaging had been
condition (DVT). Various in signs and also symptoms of widely implemented as it is noninvasive method and its low
DVT cause the diagnosis of DVT is challenging. In cost that has allowed the use of this technique for more
addition, the thrombus that develop may be asymptomatic. clinical studies. The B-mode image obtain had been
That increasingly causes the difficulty in determining the analysed to determine the condition of the patient. In this
DVT. Therefore, appropriate conduct to deal with properties study, the purpose in analysing the image using the edge
of DVT is necessary. Among these methods applied, detection is to measure the vessel wall elasticity of the
ultrasonography is as accurate as any, with more advantages popliteal vein. Moreover, several methods of edge detection
than CT, MRI, plethysmography and D-dimer test, had been applied to compare the vessel wall elasticity
including low cost, portability, non-invasive, and simplicity. measurement of the popliteal vein. Generally, edge is the
Thus, ultrasound technology is employed, which shows a plot of high intensity pixel and its immediate
number of successful DVT diagnosis (Stephanie Sippel). neighbourhood in the images. Therefore, the shape of the
Since Deep Vein Thrombosis (DVT) can give serious threat image object decided by its edges. The edges are used in
to the health condition, it is essential to diagnose the DVT the image analysis to discover its region boundaries. Edge
condition at the earliest stage. The diagnosis at this level detection is a significant process in computer vision and
could help in treating or improving the condition of patients image processing, thus, it is essential to consider the edge
with the DVT. The most common diagnoses done by the detection operators (Jie Yanga, 2008, P. P. Acharjya, 2012).
physicians are using the clinical assessment. However, the Rapid change in the pixel intensity of the image is the
result shown that the condition of the patient still cannot be common meaning of edge. It includes the important features
confirmed ( Juan N. Useche, 2008). and critical characteristics of an image. Those rapid changes
Therefore, analysis of the ultrasound image plays a in the pixel intensity, detected by using the first and second
crucial role in the medical field. Sanjay et al. in his study order derivatives. The edge is the boundary between the
reported analysis of Doppler ultrasound finding for DVT object and its background. The edge detector aimed to
(Sanjay M. Khaladkar, 2014 ). In their study, the criteria avoid the false edges and correctly detecting the true edges
that had been considered include the visualization of ( Pushpajit A. Khaire, 2012).
thrombus, vein compressibility, vein size and also The edge detection technique had been applied to
respiratory changes in the subjects. Based on the analysis, the B-mode image obtained as to analyse the characteristics
the predominant distribution of thrombus developed to be in of the popliteal vein. In edge detection, there are several
the above knee region that is the superficial femoral vein methods that had been widely used for analysis, such as
with 88.5%, while for the popliteal vein region involved Sobel, Prewitt, Canny, Roberts, LOG and etc.
with 69.2%. In the study, they concluded that colour
 VOL. X, NO. X, XXXXXXXX ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2013 Asian Research Publishing Network (ARPN). All rights reserved.

Therefore, in this study, several edge detection

methods had been combined as to compare the output result
obtained from the analysis. Each algorithm of the edge
detection method had its own specified approach, thus, the Popliteal Valve
result may vary for each method. Since then, these edge vein
detection operators are differ in term of their localization,
derivatives and magnitude (D. Poobathy, 2014). The result
obtained will be measured based on the noise level Popliteal
measurement. Noise level, there were two metric standards
artery
that universally followed, Means Squared Error (MSE), and
Peak Signal to Noise Ratio (PSNR). Commonly, the MSE
and PSNR are used in Image Compression. Then, here
both measurements were used to compare the edge
detected image quality. The MSE represents the cumulative
squared error between the edge detected and the original
image, whilst the PSNR represent the measure of the peak Figure 1: The B-mode image of the valve in the popliteal
error. Moreover, in comparison of both original image and vein, the popliteal vein positioned side by side to the
the edge detection operators, if the image obtains from popliteal artery (N. Harun, 2014).
the operator give more PSNR and low MSE, it concludes
that the operator had good image quality (D. Poobathy,
2014). C. Mean squared error (MSE)
The MSE had been included as of the degradation
METHOD function and statistical characteristics of noise on the edge
A. Experimental Procedure detected image. It is the purpose of measuring the average
To study the vessel condition of the subject for squared difference between the estimator and the parameter.
early diagnosis of DVT, experiment on 10 volunteered Besides, MSE specifies the average differences of the pixel
subjects had been done. The equipment used for the throughout the original ground truth image with the edge
experiment is TOSHIBA SSA-580A of the ultrasound detected image. The higher in value of MSE indicates that a
machine. For the scanning of the leg used the linear greater difference between the original and processed image
transducer with frequency range of 6 MHz to 12 MHz. (P. Vidya). Equation 1 shows the relationship between the
Besides, narrower transducers may make it harder to localize original image and the edge detected image for the MSE.
the veins and to apply uniform compression. Therefore, for
,
 , ,
MSE 
larger patients, a lower frequency or even an abdominal
,
probe will facilitate greater tissue penetration. The patient (1)
can be placed in either a prone position, or seated on the
edge of the bed with the knee flexed and the foot supported. where
In the prone positions the leg being examined should be M,N : size of image
down.Video of the scanning is recorded directly from the x : height of image
ultrasound machine in the range of 30 seconds to 60 y : width of image
seconds, using external frame grabber DVI2USB 3.0. This  ,  : original image
video is recorded for the measurement of the vein wall  ,  : edge detected image
displacement (N. Harun, 2014).
The MSE should be less when deal with image
B. Region Of Interest (ROI) restoration, reconstruction and compression. Thus, in terms
In the study, the region of interest (ROI) is focused on of edge detection, the MSE could be lower to ensure more
the area of valve in the popliteal vein. The ROI chosen edge points are found in the image and able to detect the
due to the necessity of doing the measurement of the weak edge points (D. Poobathy, 2014).
displacement between the popliteal vein wall. The patent of
valve movement shows the opening and closing of valve D. Peak signal to noise ratio (PSNR)
during the blood flow. Therefore, by having the valve PSNR is the ratio between the possible maximum power
scanning area, it will allow to measure the significant of a signal and the power of the corrupting noise that affect
changes of the displacement in between the wall of the the fidelity of its representation (D. Poobathy). PSNR is the
popliteal vein. The measurement taken for the wall estimation to human perception of reconstruction image
displacement is at the location near the valve area (N. quality (P. Kellman). Generally, the higher PSNR normally
Harun, 2014). Figure 1 shows the B-mode image of the indicates the reconstruction of higher quality image
popliteal vein. compression, The PSNR usually expresses in decibel (dB)
scale. Equation 2 shows the PSNR calculation based on the
MSE.

%
PSNR  10 log# $ (
&'
(2)
 VOL. X, NO. X, XXXXXXXX ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2013 Asian Research Publishing Network (ARPN). All rights reserved.

R is the maximal variation in the input image data.

E. Image Processing
Edges are detected by using the Sobel, Canny, Roberts,
combination of Canny and Sobel, and also combination
method of Canny and Roberts. For Sobel and Roberts
operators, the method applied by thresholding the gradient
function (Mamta Juneja,2009). While for the Canny
method, a threshold is applied to the gradient using the
derivatives of Gaussian filter. Figure 2 shows an example of
the original image used for the edge detection method.

Region of interest for the

edge detection Figure 3: Sobel edge map of Figure 2.

ii. Detection using Roberts operator.

Roberts cross calculates an efficient, simple, 2-D spatial
gradient measurement on an image that highlighting the
regions corresponding to the edges. The Roberts operator is
applied using two convolution mask/kernel (O. Marques,
2011). In most common usage, Roberts operator use input
of grayscale image as well as the output produce. Pixel
values at each point in the output represent the estimated
absolute magnitude of the spatial gradient of the input
image at that point. Equation (5) and (6) show the 2x2
convolution kernels of the operator.

0 ,1
Figure 2: The original image of the popliteal vein (B-mode
ultrasound image). 0*  1 2
1 0
(5)

,1 0
i. Detection using the Sobel operator.
0/  1 2
0 1
Sobel filter has a simple approximation concept of (6)
gradient with smoothing. The 3x3 convolution mask is
usually realized to detect gradients in X and Y directions.
Equation (3) and (4) show the operator that consists of a
pair of 3x3 convolution masks (kernels) (O. Marques, 2011).
Figure 3 shows the example of an image using the Sobel
operator.

,1 0 1
)*  +,2 0 2. (3)
,1 0 1

,1 ,2 ,1
)/  + 0 0 0. (4)
1 2 1
Figure 4: Roberts edge map of Figure 2.

Figure 4 shows the example of image using the Roberts

operator.

iii. Detection using Canny.

The Canny edge detector is one of the most popular
and effective edge detection operators that available (O.
Marques, 2011). The Canny method finds edges by looking
for local maxima of the gradient of the image. The
method uses two thresholds to detect strong and weak
edges (Mamta Juneja, 2009).
 VOL. X, NO. X, XXXXXXXX ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2013 Asian Research Publishing Network (ARPN). All rights reserved.

Figure 5 shows the example image of the Canny edge

method.

Figure 7: Combination of Canny and Roberts edge map of

Figure 2.
Figure 5: Canny edge map of Figure 2.

iv. Combination of Canny with Sobel and Canny with PERFORMANCE EVALUATION
A. MSE and PSNR measurement
Roberts.
As mentioned earlier, the edge detection methods were
In the combination of Canny with Sobel and Canny
assessed by objective quality measure to find reliable
with Roberts operators, the performance evaluation
indicated by the measurement of MSE and PSNR. Thus, statistical evidence in distinguishing among the output
images obtained. The numerical results shown in Table 1 and
the quality of the output image from both combinations can
Table 2 were the outcomes from the Matlab program which is
be compared and determine the acceptable method for edge
developed to process the B-mode images and to compute
detection in vessel wall elasticity measurement. Figure 6
and Figure 7 shows the example of image process by the MSE and PSNR value of the output images. The MSE
both combinations Canny with Sobel and Canny with values measured in the range of 1.60 to 16.73. The PSNR
value measured at range start from 5.9 dB to 15.98 dB. Based
Roberts operators.
on the measurement obtained, there were an absolute
difference in the measurement of MSE and PSNR for
different edge detection operators.Figure 8 and Figure 9 show
the comparison of average value between the applied
algorithm in the measurement of MSE and PSNR calculation.
It shows the pattern of the value for the mean MSE and the
mean PSNR for each edge detection method applied.
Significantly, from figure 8 and figure 9, the
combination of Canny and Sobel operator had the lowest
value of PSNR which is 9.62 dB and with the highest average
value of MSE that is 7.68. While for the Roberts operator, it
has the highest average value of PSNR that is 12.82 dB and
lowest average range of MSE which is 4.49. Moreover,
based on the output image observed for the Sobel and
Roberts operators, the operators were not acceptable for the
Figure 6: Combination of Canny and Sobel edge map of analysis of edge detection for the vessel wall measurement,
Figure 2. eventhough that both operators having the higher value of
PSNR indicate that the image is good in image quality. The
resultant images of the Sobel and Roberts did not allow for
the wall measurement because the vessel wall of the vein
cannot be seen clearly and the measurement cannot be done
based on that images. Therefore, for the image analysis of the
B-mode ultrasound image edge detection using the
combination of the canny and Robert is the best option since
the PSNR value is high and the MSE is low indicates the good
quality of image produced.
 VOL. X, NO. X, XXXXXXXX ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2013 Asian Research Publishing Network (ARPN). All rights reserved.

Table 1: Mean MSE of different edge detection in the B-mode image popliteal vein (for 10 frames) for each subjectsa.

SUBJECT Canny Roberts Sobel Canny+Roberts Canny+Sobel

1 3.909 2.120 3.057 3.572 5.545
2 7.351 5.888 7.165 7.043 7.101
3 3.864 2.008 2.589 3.807 5.844
4 7.054 5.684 6.215 6.976 8.335
5 4.501 2.120 2.789 4.029 6.481
6 5.962 2.856 3.415 5.550 8.114
7 2.359 1.640 1.870 2.389 3.579
8 16.433 15.981 15.743 16.004 16.731
9 4.422 2.095 2.700 4.207 6.393
10 6.650 4.493 4.566 6.235 8.638
a
Data are presented as mean value for each measurement of subject.

Table 2: Mean PSNR (dB) of different edge detection in the B-mode image popliteal vein (for 10 frames) for each subjectsa .

SUBJECT Canny Roberts Sobel Canny+Roberts Canny+Sobel

1 12.29 14.87 13.29 12.60 10.70
2 9.46 10.43 9.59 9.65 9.62
3 12.30 15.23 14.01 12.39 10.48
4 9.65 10.60 10.20 9.70 8.93
5 11.60 14.88 13.68 12.10 10.02
6 10.38 13.57 12.80 10.69 9.04
7 14.50 15.98 15.42 14.40 12.69
8 5.98 6.10 6.16 6.09 5.90
9 11.68 14.93 13.75 11.90 10.08
10 9.90 11.60 11.54 10.18 8.77
a
Data are presented as mean value for each measurement of subject.

9 14 12.82
7.68 12.04
8 10.97
12 10.77
7 6.25 5.98 9.62
6 10
5.01
PSNR (dB)

5 4.49 8
MSE

4 6
3
4
2
1 2
0 0

Figure 8 : Average value of MSE of edge detector. Figure 9 : Average value of PSNR of edge detector.
 VOL. X, NO. X, XXXXXXXX ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2013 Asian Research Publishing Network (ARPN). All rights reserved.

B. Vessel wall elasticity Based on this equation, the strain, F measurement is the
From the B-mode ultrasound image, the edge detector changes of the vessel wall displacement. Therefore, the
algorithm chosen was applied. The edge detection edge detection of the vessel wall which is for the wall
algorithms chosen was the combination of Canny and displacement measurement is crucial for this particular
Robert’s algorithm since it has the smallest MSE and purpose of the elasticity measurement. The elasticity of the
largest value of PSNR. This edge detection implementation blood vessel is an important parameter for the early
to the B-mode ultrasound image is important for the detection of the Deep Vein Thrombosis condition. A
measurement of the wall displacement of the popliteal vein number of subjects will be examined using the ultrasound
because of the requirement to measure the elasticity of the and the assessement of the vessel wall elasticity for the
blood vessel. Based on the Hooke’s Law anology, which subjects will be done. Thus, the measurement of the vessel
describes the relationship between stress and strain (Walter F. wall elasticity of the subjects will be compared to the
Boron, 2008): vessel wall elasticity of the subject with Deep Vein
Thrombosis condition.
=
&45677 ; BC?
E Pa  &4589:  <  @A@??  D
Based on Figure 10, shows an example of the
,
> vessel wall elasticity measurement for a subject with no
? ∆C
(7)
@?
history of Deep Vein Thrombosis with the implementation
where of the combination Canny and Robert’s algorithm. Based
on the result obtained, the average range of the strain and
E : Young's modulus (modulus of elasticity) elasticity measurement were 1.05% and 0.420MPa
respectively. In addition, for the future development of this
F : Force exerted on an object under tension
study, more subjects will be examined to differentiate the
A0: Original cross-sectional area through which the force difference of the vessel wall elasticity measurement for the
is applied subject with no history of Deep Vein Thrombosis and the
∆L: Amount by which the length of the object changes subject that has Deep Vein Thrombosis condition.
L0 : Original length of the object

Strain (%)
3.5 Vessel Wall Elasticity Elasticity (MPa)
3.5
Strain (%)
3 3
Elasticity
2.5 (MPa) 2.5

2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Time (s)
Figure 10 : Edge detection Canny with Roberts for vessel wall elasticity measurement.

CONCLUSION future, more alternatives in analyzing the ultrasound image

This paper discussed five different resultant images can be present to realize the efficiency of the image analysis
of five condition edge detection algorithms. The evaluation for early detection of Deep Vein Thrombosis condition.
of the performance measured based on the Mean Squared
Error (MSE) and also the Peak Signal to Noise Ratio ACKNOWLEDGEMENT
(PSNR). The results show that image with higher levels of The authors like to thank Research Acculturation
PSNR and low level of MSE, produced a better image Grant Scheme (RAGS: no. of R031) for the sponsorship
visualization for analysis. As a result, the image combination doing this project.They also gratefully acknowledged the
of Canny and Robert’s edge detection method produced an support and generosity from Universiti Tun Hussein Onn
approximate value of vessel wall elasticity measurement at Malaysia (UTHM), Medical Imaging in Laboratory Electrical
the popliteal vein. Therefore, the combination of the Canny and Electronic engineering faculty and Microelectronics and
and Robert’s edge detection method can be an alternative in Nanotechnology-Shamsuddin Research Centre (MiNT- SRC)
analysing the B-mode the ultrasound image instead of for having the space doing the experiment and analysis for
measurement from the raw image of ultrasound. In the near this project.
 VOL. X, NO. X, XXXXXXXX ISSN 1819-6608
ARPN Journal of Engineering and Applied Sciences
©2006-2013 Asian Research Publishing Network (ARPN). All rights reserved.

REFERENCES
D. Poobathy and R. Chezian, “Edge Detection Operators: Walter F. Boron, Emile L. Boulpaep, 2008, ‘Elastic
Peak Signal to Noise Ratio Based Comparison”, I. J. Image, Properties Of Blood Vessels’, Medical Physiology, Saunders,
Graph. Signal Processing, no 10. September, pp. 55–61, ISBN-10: 1416031154 ISBN-13: 978-1416031154.
2014.

Jie Yanga, Ran Yanga, Shigao Lib, S.Shoujing Yina, and

Qianqing Qina, “A Novel Edge Detection Based
Segmentation Algorithm for Polarimetric Sar Images”, The
International Archives of the Photogrammetry, Remote
sensing and Spatial Information, Sciences. Vol. XXXVII,
Part B7.Beijing, 2008.

J. N. Useche, A. M. F. de Castro, G. E. Galvis, R. A.

Mantilla, and A. Ariza, “Use of US in the evaluation of
patients with symptoms of deep venous thrombosis of the
lower extremities.,” Radiographics, vol. 28, no. 6, pp. 1785–
1797, 2008.

M. Juneja and P. S. Sandhu, “Performance evaluation of edge

detection techniques for images in spatial domain,” Int. J.
Comput. Theory Eng., vol. 1, no. 5, pp. 614–621, 2009.

N. Hafizzatul, M. Harun, N. Ibrahim, W. Nurshazwani, W.

Zakaria, N. Farhan, and N. Shazilah, “Study of Vessel
Conditions in Different Categories of Weight for Early-Stage
of Deep Vein Thrombosis ( DVT ) Diagnosis,” vol. 6, no. 3,
pp. 57–64, 2014.

O. Marques, “Practical Image and Video Processing Using

MATLAB”, IEEE Press, John Wiley & Sons, Inc, Hoboken,
New Jersey, pp. 335-364, 2011.

Peter Kellman and Elliot R. McVeigh, “Image

Reconstruction in SNR Units: A General Method for SNR
Measurement”, Magn Reson Med. Author manuscript, Magn
Reson Med. Volume 54(6) pp. 1439–1447, December 2005.

Pinaki Pratim Acharjya, Ritaban Das and Dibyendu Ghoshal,

“Study and Comparison of Different Edge Detectors for
Image Segmentation”, Global Journal of Computer Science
and Technology Graphics & Vision, Volume 12 Issue 13, pp.
28- 32, 2012.

Pushpajit A. Khaire and Dr. N. V. Thakur, “A Fuzzy Set

Approach for Edge Detection”, International Journal of
Image Processing (IJIP), Volume 6 Issue 6, pp. 403-412,
2012.

P. Vidya, S. Veni and K.A. Narayanankutty, “Performance

Analysis of Edge Detection Methods on Hexagonal Sampling
Grid”, International Journal of Electronic Engineering
Research, Volume 1, pp. 313–328, Number 4 2009.

S. Khaladkar, D. Thakkar, K. Shinde, D. Thakkar, H. Shrotri,

and V. Kulkarni, “Deep vein thrombosis of the lower limbs:
A retrospective analysis of doppler ultrasound findings,”
Med. J. Dr. D.Y. Patil Univ., vol. 7, no. 5, p. 612, 2014.

Stephanie Sippel, Krithika Muruganandan, Adam Levine and

Sachita Shah, 2011, „Review Article: Use Of Ultrasound In
The Developing World‟, International Journal Of Emergency
Medicine 2011, 4:72, doi:10.1186/1865-1380-4-72 .