Improvement in diagnosis of breast tumour using ultrasound elastography and echography: A phantom based analysis
1 Department of Electrical Engineering, IIT Madras, Chennai, India
2 Department of Chemistry, IIT Madras, Chennai, India
3 Mediscan Systems, Chennai, India
Due to the isoechoic nature of lesions and their poor
contrast with neighbouring tissue, a lesion may remain undetected in ultrasound
B mode imaging for cancerous tissue. Imaging of the elastic properties of
tissue provides new information which is collateral to tissue pathology. This
study provides quantitative analysis of improvements in tumour diagnosis when
the ultrasound B mode imaging is combined with elastography. Quantification was
based on the textural parameters measured from the ultrasound B mode image and
strain measured from the elastogram. The ability of a parameter to discriminate
between diseased cases and normal cases was evaluated using receiver operating
characteristic (ROC) analysis. Polyacrylamide gel based tissue mimicking
phantoms with embedded inclusions of varying stiffness were used for the
analysis. � 2009 Biomedical Imaging and Intervention Journal. All rights
Keywords: Elastography; classifier; ROC; texture; tissue
Particularly when the task at hand is to differentiate between
malignant tumours and other possibly benign conditions, non-invasive
identification of a breast tumour is a challenging area of medical diagnosis.
Prior to the advent of diagnostic imaging, the oldest method of detection
involved palpation. While palpation is understandably simple, it is
nevertheless just a qualitative assessment and can only be applied to
superficial organs. Malignant tumours feel harder than benign ones and this
physical property is related to their coefficient of elasticity. The elasticity
of soft tissues depends on their molecular building blocks (fat, collagen,
etc.) and on their microscopic and macroscopic structural organisation of these
blocks [1-3]. Pathological changes are generally correlated with changes in
tissue stiffness as well. In many cases, the lesion may not possess sufficient
echo graphic properties and therefore it avoids detection using normal B mode
ultrasound. As the echogenecity and the stiffness of tissue are generally
uncorrelated, it is expected that imaging tissue stiffness will provide new
information that is related to tissue structure and or pathology .
Elastography is a method of estimating the elastic
properties of biological tissues [3-7]. Elastograms are obtained by comparing
ultrasound echo data obtained before and after a slight compression of the
tissue. The results of the comparison are displayed as an image, where soft
areas appear white and hard areas appear black. Garra et al.  used
ultrasound elastography for the examination of breast lesions using a 5 MHz
linear array transducer. The elastogram and corresponding sonogram were
evaluated by a single observer for lesion visualisation, relative brightness
and margin definition. The mean depth and width of the lesion on the elastogram
and sonogram were measured and used as parameter for differentiating benign and
malignant lesions. Andrej et al.  and Rago et al.  performed real time
elastography on the thyroid lesions. All elastograms were evaluated for the
lesion visibility, relative brightness, and margin regularity and definition by
using a four point scale. Zhi et al.  evaluated the capability of
ultrasound elastography in differentiating between benign and malignant lesions
of the breast, and compared this with conventional sonography and mammography.
They concluded that ultrasound elastography is superior to conventional
sonography, and furthermore is superior or equal to mammography in
differentiating between benign and malignant lesions in the breast. By
combining ultrasound elastography and sonography, accuracy of detection can be
improved greatly, while retaining the potential to reduce unnecessary biopsy.
None of the above studies attempted a quantitative
analysis of the improvement in classification accuracy when ultrasound
elastography is combined with echography. The main focus of this study is to
quantify the improvement in diagnosis of tumours by combining the ultrasound B
mode imaging with elastography. Quantification is based on the textural
parameters measured from the ultrasound B mode image and strain measured from
Materials and methods
Polyacrylamide gel based tissue mimicking phantoms 
were designed for study and analysis. The polymerisation of acrylamide was
carried out in deionised water at room temperature using bisacrylamide as the
cross linker, and mmonium persulfate (APS) and tetramethylethylenediamine
(TEMED) as a pair of redox initiators. Fine particles of titanium dioxide were
used to control the echogenecity of the gel. The elasticity of the phantom was
varied by adjusting the acrylamide concentration so that it covered a wide
range of normal to pathological tissue stiffness.
Acoustic properties of the phantoms were measured using
the pulse-echo-transmit-receive principle, with a single element transducer of
5 MHz centre frequency. Young�s modulus of the sample was measured using the
unconfined compression test. A variable compression load was applied on the
sample and corresponding deformation is measured. From these data,
stress-strain characteristics of the sample were plotted and the modulus was
Two categories of phantoms were designed with embedded
inclusions of varying stiffness. In the first category the elasticity was
maintained constant and echogenecity was varied, thus producing isoechoic
(indistinguishable from surrounding tissue), hyperechoic (appearing brighter
than surrounding tissue), hypoechoic (appearing darker than surrounding tissue)
and anechoic lesions (those without echoes). In the second category
echogenecity was kept constant and elasticity was varied.
The plan and elevation of a typical phantom from the
second category is shown in Figure 1. The external view of the phantom is shown
in Figure 2. The ultrasound B mode and elastographic behaviour of these lesions
was analysed using image processing methods.
Ultrasound B mode and elastograms were acquired using
Siemens ACUSON Antares , at Mediscan System, Chennai. Linear array
transducer VF 7-3 with a probe frequency of 5 MHz was used. Initially, B mode
image of the region of interest (ROI) was obtained and then a slight
compression was applied by pressing the transducer on the surface. Built- in
software present in the system generated the elastogram by comparing the
pre-compressed and post-compressed RF signals, and displayed these adjacent to
the B mode image for comparison and more accurate diagnosis.
Tissue characterising parameters
Texture parameters were measured to quantify the B mode
images [14-15]. The first order parameters used in this study were mean value,
variance, skewness, kurtosis, energy and entropy, and accounted for variations
in intensity without spatial reference. These parameters were measured by
computing the image histogram, the shape of which provided many clues as to the
character of the image. Second order texture parameters were extracted from the
normalised co-occurrence matrices  and include angular second momentum,
correlation, inertia, absolute value, inverse difference, entropy and maximum
probability. While first order texture parameters cannot completely
characterise texture, second order texture parameters account for changes in
the spatial arrangements of intensities.
Quantitative analysis of parameters
The ability of a test to discriminate between diseased
cases and normal cases was based on receiver operating characteristic (ROC)
analysis. The ROC curve represents a plot describing the classifier�s true
positive (TP) detection rate against its false positive (FP) rate [17-18]. In
medical imaging the TP rate is commonly referred to as sensitivity and (1.0 �
FP rate) is referred to as specificity.
The histograms of each parameter were calculated to obtain
the distribution of parameter values for both normal and tumour groups. Specificity
(Sp) and sensitivity (Se), were determined as:
where TN is true negative and FN is false
By changing the threshold, TP and FP rate can be varied.
Plotting sensitivity against specificity, for different threshold values, ROC
curve can be obtained. The area under the ROC curve (AROC) is an
accepted method of comparing classifier performance, and a perfect classifier
ought to have a TP rate of 1.0 and a FP rate of 0.0, resulting in AROC
of 1.0. Values of �indicate the potential
of the respective parameters to differentiate between the existence and absence
The probability that a sample belongs to class Ci,
given that it has a feature value x denoted by P(Ci/x) can be
computed as :
where p(x/Ci) is the conditional probability of
obtaining feature value x given that sample is from class Ci, P(Ci)
is the prior probability that a random sample is a member of class Ci
and C is the total number of classes. For a two dimensional case, the normal
density function can be written as:
where Cv is the covariance matrix, μ is
the mean vector, |Cv| and Cv-1 are the
determinant and inverse of covariance matrix respectively.
A Bayesian classifier can be trained by determining the
mean vector, and the covariance matrices for the normal and tumour classes from
the training data. For the training corresponding to Class 1 (tumour) and Class
2 (normal) data, mean vector and covariance matrix were calculated separately.
The parameters with highest values of AROC were chosen as the best
performing parameters. Using these parameters, a discriminant function for a
two class case can be defined as:
where g1(x) and g2(x) are
discriminant functions corresponding to two classes. We use the following
decision rule: decide class 1 if g(x) > 0; otherwise decide class 2. 
The acoustic properties of the phantoms used for the
analysis are shown in the Table 1 along with corresponding parameters for human
tissue [20-21]. Figures 3, 4 and 5 show the ultrasound B mode and elastogram of
the different lesions analysed.
They are hyperechoic (Figure 3), anechoic (Figure 4b) and
isoechoic (Figure 5) lesions respectively. In the elastogram , soft areas are
represented by white and hard areas are represented by black. Figure 6 shows
the areas on the B mode image and the elastogram for the measurement of
From the designed phantoms, a total of 28 images from the
embedded inclusions were obtained. Separation of normal regions and tumour
regions was based on the Young�s modulus of the inclusion, which in turn was
controlled by the concentration of monomers. An inclusion with a Young�s
modulus of greater than 60 kPa was considered as a stiffer region and therefore
deemed to be a tumour. Based on this criterion, among 28 inclusions, 7 were
selected as normal, and the remaining 21 as tumour regions.
ROC analysis was performed for all 14 parameters, and
finally parameters with AROC > 0.5 were used as input to the
classifier. From the group of first order texture parameters, mean (M), skewness
(SK) and entropy (ENT) were used. The parameters used from the second order
texture parameters were angular second momentum (ASM) and maximum probability
(max_prob). In addition, average strain (STR) estimated from the elastogram was
Results of the parameter evaluation using AROC
are shown in Table 2. The total number of cases used for training was 5 normal
and 8 tumours. Classification results with and without the average strain
parameter (STR) are shown in the Table 3.
The acoustic parameters of the phantoms designed in this
study closely matched those of human tissue (Table 1). The lesions were clearly
distinguishable in the elastogram, even though they appear almost similar in B mode.
These results are clearly evident from the images shown in Figures 3, 4 and 5.
In this study, along with average strain estimated from the elastogram, several
tissue characterising parameters were used, all estimated from ultrasound B
mode images. The ability of each parameter in differentiating between normal
regions and tumour regions has been analysed. Class 1 included lesions whose
Young�s modulus is greater than 60 kPa; Class 2 included softer regions.
ROC analysis confirmed that the use of a single parameter
is not recommended. And it has been shown that using high performance
parameters collectively in the classification will produce an accuracy of
Echogenicity and elasticity of the phantoms are suitably
controlled to obtain normal and pathological lesions in the phantoms. An
increase in the classification accuracy was achieved following the inclusion of
the average strain parameter as an additional input to the classifier.
Elastography may thus be considered as one of the available diagnosis tool for
improving the accuracy of the diagnosis of breast and thyroid tumours.
Figure 1 Plan and elevation of a typical phantom with six embedded inclusions of varying stiffness. (Phantom details: prepared with 8% acrylamide, 0.06% initiator and 40% TiO2 Concentration of acrylamide in inclusions was varied from 6% to 20%. Concentration of TiO2 and initiator was same in all the inclusions and is equal to 40% and 0.06% respectively). All measurements are in mm.
Figure 2 External view of the phantom.
Figure 3 Ultrasound B mode image and elastogram of two solid lesions. Both the lesions appear hyperechoic in B mode but their characteristics are different in elastogram. (a) is less stiffer than (b). In elastogram black region represents hard (HD) and white region represents soft (SF). Line markings shows the width and depth of the lesion.
Figure 4 Ultrasound B mode image and elastogram of two lesions. (a) Very small hyperechoic lesion in B mode and hard lesion in elastogram. (b) Anechoic lesion in B mode and a Bull�s eye appearance in elastogram, a typical elastographic appearance of cystic lesion.
Figure 5 Ultrasound B mode image and elastogram of three solid lesions. Lesions appear isoechoic in B mode but their characteristics are different in elastogram. (a) and (c) are stiffer and (b) is soft. Arrow markings show the width and depth of the lesion.
Figure 6 Ultrasound B mode image (B) and elastogram (E) of a hyperechoic lesion. Rectangular region marked in red (dashed line) in B represents the lesion in B mode image selected for parameter measurement and region marked in blue (solid line) is the region in surroundings selected for parameter measurement. Rectangular region marked in red (dashed line) in E represents the lesion in elastogram selected for average strain measurement and region marked in blue (solid line) is the region in surroundings selected for average strain measurement.
Table 1 Acoustic parameters of the designed phantom with corresponding parameters for human tissue ,  for comparison.
Table 2 Performance of selected parameters
Table 3 Classification accuracy with and without strain parameter
Fung YC. Biomechanics: mechanical Properties of Living Tissues. 2nd edition. Springer, 1993.
Yamada H. Strength of Biological Materials. Baltimore: The Williams and Wilkins Company, 1970.
Ophir J, Cespedes I. Elastography: ultrasonic imaging of tissue strain and elastic modulus in vivo. European J Med Ultrasound 1996; 3:49-70.
Hall TJ. AAPM/RSNA physics tutorial for residents: topics in US: beyond the basics: elasticity imaging with US. Radiographics 2003; 23(6):1657-71.
Ophir J, Cespedes I, Ponnekanti H et al. Elastography: a quantitative method for imaging the elasticity of biological tissues. Ultrason Imaging 1991; 13(2):111-34.
Ophir J, Alam SK, Garra B et al. Elastography: ultrasonic estimation and imaging of the elastic properties of tissues. Proc Inst Mech Eng H 1999; 213(3):203-33.
Szabo TL. Diagnostic ultrasound Imaging Inside out. Elsevier Academic Press, 2004.
Garra BS, Cespedes EI, Ophir J et al. Elastography of breast lesions: initial clinical results. Radiology 1997; 202(1):79-86.
Lyshchik A, Higashi T, Asato R et al. Thyroid gland tumor diagnosis at US elastography. Radiology 2005; 237(1):202-11.
Rago T, Santini F, Scutari M et al. Elastography: new developments in ultrasound for predicting malignancy in thyroid nodules. J Clin Endocrinol Metab 2007; 92(8):2917-22.
Zhi H, Ou B, Luo BM et al. Comparison of ultrasound elastography, mammography, and sonography in the diagnosis of solid breast lesions. J Ultrasound Med 2007; 26(6):807-15.
Kumar K, Jayashankar V. Image processing based characterization of tissue phantoms for ultrasound elastography applications. Proc Int Conf Comp Comm Control and Instru. 2007: 395-9.
Siemen�s Medical [Online]. Available at http://www.medical.siemens.com.
Wu CM, Chen YC, Hsieh KS. Texture features for classification of ultrasonic liver images. IEEE Trans Med Imaging 1992; 11(2):141-52.
Garra BS, Krasner BH, Horii SC et al. Improving the distinction between benign and malignant breast lesions: the value of sonographic texture analysis. Ultrason Imaging 1993; 15(4):267-85.
Materka A, Strzelecki M. Texture analysis methods - a review. Brussels: Technical university of Lodz, Institute of electronics, 1998; COST B11 report.
Meyer-Base A. Pattern recognition for medical imaging. Elsevier Academic Press, 2004.
Joo S, Yang YS, Moon WK et al. Computer-aided diagnosis of solid breast nodules: use of an artificial neural network based on multiple sonographic features. IEEE Trans Med Imaging 2004; 23(10):1292-300.
O�Duda R, Peter EH, David GS. Pattern classification. 2nd edition. John Wiley and Sons, 2001.
Hill CR, Bamber JC, ter Haar GR. Physical Principles of Medical Ultrasonics. John Wiley and Sons, 2004.
George DL. The velocity of sound through tissues and the acoustic impedance of tissues. J Acoust Soc Am 1950; 862-6.
|Received 7 May 2009; accepted 18 September 2009
Correspondence: c/o Prof. Jayashankar V, ESB 317A, Measurements and Instrumentation Lab, Department of Electrical Engineering, IIT Madras, Chennai � 600036, India. Tel.: +91-44 -2257 5404; Fax: +91-44-2257 0509; E-mail: firstname.lastname@example.org (Kishore Kumar).
Please cite as: Kumar K, Andrews ME, Jayashankar V, Mishra AK, Suresh S,
Improvement in diagnosis of breast tumour using ultrasound elastography and echography: A phantom based analysis, Biomed Imaging Interv J 2009; 5(4):e30