Key words
breast - elastography - strain ratio - ultrasound - specificity
Introduction
Over the last 20 years, sonoelastography has developed from a technically complex
examination method to one that is simple to carry out and reproduce, and that can
be integrated into clinical examination procedures [1]. Various manufacturers of ultrasound devices have integrated elastography as a standard
feature, and its use in a range of clinical settings is the subject of numerous current
studies [2]
[3]
[4]
[5]. Previous studies have provided evidence for the benefits of sonoelastography as
an additional procedure in breast ultrasound [6]
[7]
[8]
[9]
[10]
[11]. A 5-stage elasticity score has been established by Ueno et al. as the standard
in elastography evaluation [12]. However, the use of this score requires a certain amount of experience; the interobserver
and intraobserver variability is considerable [13]. The calculation of the strain ratio is a semi-quantitative measurement that indicates
the average deformation of a lesion in relation to the deformation in a reference
region (usually the lateral, surrounding fatty tissue) [14]. The measurement results in a metric value with no specified dimensions. More recent
studies have shown a significant difference between the average strain ratio of benign
and malignant breast tumors [6]
[15]
[16]
[17]
[18].
The aim of our study was to determine the value of strain ratio to discriminate between
malignant and benign breast tumors in comparison to the qualitative 5-stage elasticity
score and conventional B-mode ultrasound. Secondary study aim was to investigate factors
that limit the use of sonoelastography.
Materials and Methods
Patients
Between January 2010 and October 2010, a total of 215 patients with 224 breast masses
were prospectively enrolled in the study. B-mode ultrasound was carried out on all
patients, and mammograms were available for 190 patients. Mammography was not carried
out on 25 patients due to their young age. After informed consent, real-time elastographic
examination was carried out and the data sets were saved. Final diagnoses were made
on the basis of histological findings. The local institutional review board approved
the study.
Data Acquisition/Imaging Methods
Conventional ultrasound images and real-time elastographic data sets were obtained
from three very experienced ultrasound examiners using a 12.5-MHz linear transducer
(Philips iU22, Bothell, WA, USA). The principles of ultrasound real-time elastography
are well-known [12]
[19]
[20]. The sonoelastographic function of the Philips software is based on tissue compression
during thoracic respiratory excursions. In order to obtain appropriate images, the
transducer has to be applied with only the light pressure necessary to maintain contact
with the skin. The results are displayed as color-coded images, with blue representing
hard tissue and red to green representing soft tissue on a continuous scale ([Fig. 1]). In addition, for semi-quantitative evaluation of the strain ratio, a 60-second
cineloop was obtained during elastographic examination. The strain ratio describes
the fat to lesion ratio, indicating the stiffness of a lesion. It is obtained by setting
a rectangular Region of Interest (ROI) to cover the subcutaneous fatty tissue and
a second ROI to cover the lesion. The Philips iU22 software package features the possibility
of retrospectively analyzing the strain ratio (known as „deformation quotient”). The
cineloop can be used to create a deformation quotient curve, which can in turn be
used to calculate the mean and maximum deformation quotients ([Fig. 2]).
Fig. 1 Ultrasound and elastography images of a benign fibroadenoma. B-mode ultrasound images
showing a lobulated mass with fuzzy margins according to category 4 of BIRADS classification.
Elastography image showing high peripheral strain that corresponds to Score 3 of a
5-point elasticity score (Tsukuba score).
Abb. 1 Sonografischer Befund eines Fibroadenoms. Im B-Bild lobulierter echoarmer Herd mit
partiell unscharfer Begrenzung entsprechend einer BIRADS-4-Läsion. Das Elastogramm
zeigt eine hohe periphere Dehnbarkeit (grün) sowie zentral wenig elastische Anteile
(blau gefärbte Areale) und entspricht einem Tsukuba-Elastizitätsscore von 3.
Fig. 2 Strain ratio (SR) analysis. SR is determined by calculating the strain measured in
the adjacent fatty tissue and the strain measured in the lesion. The strain ratio
reflects the relative stiffness of a lesion. High strain ratio levels indicate an
increasing probability of malignancy.
Abb. 2 Berechnung der Strain Ratio (SR). Durch Platzierung einer rechteckigen Region of
Interest (ROI) in das umgebende Fettgewebe (= Deformierung 1) sowie einer zweiten
ROI in die Läsion (Deformierung 2) wird das Verhältnis der Elastizität beider Gewebe
(= Deformationsquotient) abgebildet. Hohe SR-Werte reflektieren eine hohe Gewebesteifigkeit
der Läsion und sind mit einer höheren Wahrscheinlichkeit eines malignen Befunds assoziiert.
All acquired data were reviewed by an experienced ultrasound examiner. The B-mode
images were evaluated according to the Breast Imaging Reporting and Data System (BIRADS)
classification. BIRADS categories 1 to 3 were assumed to be benign, and BIRADS categories
4 and 5 were taken as malignant. The elastographic data sets were analyzed qualitatively
using the Tsukuba elasticity score. This is a color-coded 5-stage score with score
1 revealing the whole lesion in green and score 5 revealing the lesion and its surrounding
tissue in blue. Lesions categorized as score 1, 2 or 3 are assumed to be benign, while
lesions categorized as score 4 or 5 are assumed to be malignant. Further details can
be found elsewhere [12].
In order to carry out a semi-quantitative elastographic evaluation, the strain ratio
of the lesions was obtained. According to the user manual, the most appropriate single
image should be selected for further analysis. Since we observed significant intraobserver
variability, the following protocol was used: All measurements (setting the ROIs)
were carried out three times. A mean was calculated from the three results for mean
deformation quotient as well as maximum deformation quotient. This procedure was carried
out with a single image, with a sequence of 20 images, and with the entire cineloop
(60 images).
Statistics
Statistical analysis was carried out in collaboration with an independent statistician
(G. K.) using the SPSS 19.0 software package (IBM Ehningen, Germany). First, descriptive
statistics were computed for variables of interest. The statistics computed included
mean, median, minimum, maximum and standard deviations of continuous variables, frequencies
and percentage frequencies of categorical factors. Additionally, 95 % confidence intervals
are presented. Furthermore, comparison of the strain ratio means between benign and
malignant masses was realized by using Student t test. In order to further evaluate the performance of strain ratio in the differentiation
between benign and malignant lesions, a Receiver Operating Characteristic (ROC) analysis
was carried out to compare the area under the curve (AUC). The best results were obtained
using the mean strain ratio of the sequence image analysis, so these data were used
for further evaluation. The diagnostic sensitivity, specificity, positive and negative
predictive values and accuracy were calculated for the qualitative elasticity score
and the strain ratio in comparison with mammography and conventional ultrasound. Finally,
we compared the diagnostic properties of the strain ratio according to lesion size,
distance from the skin (lesion depth), breast density and histological subtypes. All
p-values resulted from two-sided statistical tests and p ≤ 0.05 was considered to
be significant.
Results
Study population
The patients’ median age was 56 years, with a range of 12 to 90 years. Of the 224
lesions, 116 (51.8 %) were malignant and 108 (48.2 %) were benign. The lesion size
ranged from 4 to 110 mm with a median of 15 mm. There were 125 (55.8 %) palpable masses
among the study population. Concerning breast density, 10 % revealed fatty tissue,
47 % scattered fibroglandular tissue, 35 % heterogeneously dense tissue and 8 % extremely
dense tissue.
Strain ratio
The mean strain ratio for malignant breast tumors was 3.04 ± 0.9 (mean ± standard
deviation SD, range 0.86 to 5.43) and was significantly higher than for benign breast
tumors (1.91 ± 0.75, range 0.31 to 3.91) ([Fig. 3]). [Fig. 4] shows the diagnostic performance of the strain ratio using the ROC curve. At a cutoff
of ≤ 2.0 for benign tumors and > 2.0 for malignant tumors, sensitivity was 90.7 %
(95 % CI 83.6 %; 94.8 %), specificity was 59.2 % (95 % CI 48.3 %; 67.4 %), the positive
predictive value (PPV) was 70.3 % (95 % CI 62.2 %; 77.3 %), the negative predictive
value was 85.1 % (95 % CI 74.7 %; 91.7 %) and accuracy was 75.1 % (95 % CI 69.2; 81).
Fig. 3 Boxplot graphs demonstrate the significant difference in the strain ratio values
between malignant (n = 116) and benign (n = 108) breast lesions.
Abb. 3 Boxplot-Diagramm der Strain Ratio (SR) von malignen (n = 116) und benignen (n = 108)
Mammabefunden. Die Differenz ist hochsignifikant.
Fig. 4 Receiver Operating Characteristics (ROC) curves for the sonoelastographic strain
ratio, the Tsukuba elasticity score, B-mode ultrasound and mammography. The AUC values
did not vary significantly between the imaging methods.
Abb. 4 Diagnostische Performance der Strain Ratio, des Tsukuba-Elastizitätsscores, des B-Mode-Ultraschalls
und der Mammografie, dargestellt mittels Receiver-Operating-Characteristics(ROC)-Kurven.
Die Area-Under-the-Curve(AUC)-Werte weisen keinen signifikanten Unterschied auf.
Diagnostic Performance in Comparison to Mammography, B-mode Ultrasound and Qualitative
Sonoelastography (5-point scale)
Diagnostic Performance in Comparison to Mammography, B-mode Ultrasound and Qualitative
Sonoelastography (5-point scale)
[Table 1] shows the sensitivity, specificity, PPV, NPV and accuracy in comparison to conventional
methods (mammography and B-mode ultrasound) and to qualitative sonoelastography using
a 5-point scale (1 – 3 benign, 4 – 5 malignant).
Table 1
Diagnostic performance of strain ratio and 5-point score elastography (Tsukuba score)
vs. B-mode ultrasound and mammography.[1]
|
strain ratio
|
5-point score elastography
|
B-mode Ultrasound
|
mammography
|
|
sensitivity (%)
|
90.7 (83.6 – 94.8)
|
87.9 (80.8 – 92.7)
|
97.4 (92.7 – 99.1)
|
92.9 (86.5 – 96.3)
|
|
specificity (%)
|
58.2 (48.3 – 67.4)
|
73.1 (64.1 – 80.6)
|
42.6 (33.7 – 52)
|
56.4 (45.4 – 66.9)
|
|
PPV (%)
|
70.3 (62.2 – 77.3)
|
77.9 (70 – 84.1)
|
64.6 (57.2 – 71.3)
|
75.4 (67.6 – 81.8)
|
|
NPV (%)
|
85.1 (74.7 – 91.7)
|
84.9 (76.3 – 90.8)
|
93.9 (83.5 – 97.9)
|
84.6 (72.5 – 92)
|
|
accuracy (%)
|
75.1 (69.2 – 81)
|
80.8 (75.4 – 86.2)
|
71 (65.1 – 76.9)
|
77.9 (72 – 83.8)
|
|
AUC
|
0.83 (0.78 – 0.89)
|
0.87 (0.82 – 0.92)
|
0.82 (0.77 – 0.88)
|
0.85 (0.80 – 0.91)
|
1 Data presented as mean (95 % confidence interval).PPV = Positive Predictive Value,
NPV = Negative Predictive Value, AUC = Area Under the Curve.
Although the sensitivity of elastography is lower than for B-mode ultrasound, it provides
higher specificity. At 80.8 %, qualitative sonoelastography provides the greatest
accuracy.
If we consider the sub-group with BIRADS 4 in B-mode ultrasound, this results in a
pretest probability for a malignant tumor of 48 %; after including sonoelastography
for scale 1 – 3/strain ratio ≤ 2.0, this results in a posttest probability of 25 %/25.8 %;
and for scale 4 – 5/strain ratio > 2.0, this results in a posttest probability of
67.2 %/56.9 %.
Strain Ratio Depending on Tumor Size, Distance from Skin, Breast Density and Histological
Diagnosis
The extent to which the predictive accuracy of strain ratio depends on external factors
was investigated using bivariate correlation analysis. The analysis shows a statistically
significant negative correlation between the distance of the tumor from the surface
of the skin and the correct estimation of whether the tumor is malignant or benign
(p = 0.038). If we compare the group of tumors located close to the surface (skin
distance ≤ 4 mm) with deeper lesions (skin distance > 4 mm), the latter group displays
considerably higher sensitivity (94.2 vs. 75 %) and specificity (61.5 vs. 48.4 %).
Breast density and tumor size did not influence the diagnostic performance of the
strain ratio.
[Fig. 5] shows the average strain ratio of the various histological diagnoses. It is shown
that for malignancies, the average strain ratio is considerably higher than for benign
histologies. Particularly, this applies to invasive lobular breast carcinomas (mean
3.69; range 2.62 – 5.43; n = 10), which are often difficult to judge using mammography
and sonography. Of the benign tumors, fatty necrosis (mean 2.78; range 1.67 – 3.62;
n = 7) and radial scars (mean 2.46: range 1.22 – 3.36; n = 4) show a high strain ratio.
Fig. 5 Boxplot graphs demonstrate strain ratio values according to histologic subclassification.
DCIS = Ductal Carcinoma in Situ, ADH = Atypical Ductal Hyperplasia.
Abb. 5 Boxplot-Diagramm der SR in Abhängigkeit von der histologischen Diagnose. DCIS = duktales
Carcinoma in situ, ADH = atypische duktale Hyperplasie.
The diagnostic accuracy of the various methods (mammography, B-mode ultrasound, 5-point
scale sonoelastography and strain ratio) in the different histological categories
is shown in [Table 2]. Of the malignant tumors, mucinous breast carcinoma is not identified as malignant
by sonoelastography (5-point scale) (0/3) and is identified as malignant by strain
ratio in 2/3 cases. In identifying cysts as benign, sonoelastography outperforms B-mode
ultrasound (5-point scale elastography 9/11, strain ratio 10/11, B-mode ultrasound
5/11). In all examination methods, fat necrosis and radial scars represent a diagnostic
challenge.
Table 2
Accuracy of diagnostic properties in different histological categories.[1]
|
strain ratio
|
5-point score elastography
|
B-mode ultrasound
|
mammography
|
|
benign
|
malignant
|
benign
|
malignant
|
benign
|
malignant
|
benign
|
malignant
|
|
IDC
|
6
|
75
|
7
|
78
|
1
|
84
|
6
|
75
|
|
ILC
|
0
|
10
|
0
|
13
|
0
|
13
|
1
|
12
|
|
IMC
|
1
|
2
|
3
|
0
|
0
|
3
|
1
|
2
|
|
DCIS
|
1
|
4
|
1
|
4
|
1
|
4
|
0
|
5
|
|
lymphoma
|
1
|
1
|
1
|
1
|
1
|
1
|
0
|
2
|
|
ADH
|
1
|
0
|
3
|
0
|
1
|
2
|
3
|
0
|
|
papilloma
|
0
|
1
|
2
|
1
|
1
|
2
|
1
|
0
|
|
fibroadenoma
|
27
|
18
|
35
|
11
|
26
|
20
|
13
|
13
|
|
cyst
|
10
|
1
|
9
|
2
|
5
|
6
|
7
|
1
|
|
mastopathy
|
6
|
2
|
8
|
2
|
4
|
6
|
7
|
3
|
|
fat necrosis
|
1
|
6
|
2
|
5
|
0
|
7
|
3
|
3
|
|
radial scar
|
1
|
3
|
1
|
3
|
0
|
4
|
0
|
4
|
1 IDC = Invasive Ductal Carcinoma, ILC = Invasive Lobular Carcinoma, IMC = Invasive
Mucinous Carcinoma, DCIS = Ductal Carcinoma in Situ, ADH = Atypical Ductal Hyperplasia.
Discussion
In the presented study, the raw data sets of the elastograms saved in DICOM format
were retrospectively blinded and evaluated, and the strain ratio was calculated for
each data set. However, integration into the current examination process does not
seem to involve any problems and requires an additional time of approximately 2 – 3
minutes. This corresponds to the experiences of other working groups [21].
Previous studies have shown the potential of elastography for differentiating between
benign and malignant breast tumors [6]
[8]
[16]
[23]. Our study reveals a significant difference in the average strain ratio of benign
and malignant lesions. At a cutoff value of ≤ 2.0, both the strain ratio and the 5-point
score demonstrate a higher specificity and higher PPV than B-mode ultrasound and mammography,
albeit with a slightly lower sensitivity. In terms of accuracy, 5-point score elastography
was superior to the other examination methods. Other studies have demonstrated an
increase in NPV with additional use of elastography [7]
[18]. However, at an NPV of 85 % for both elastography methods, histological findings
for sonographically suspicious lesions must be verified. In a multi-center study by
Wojcinski et al., the following algorithm was proposed for the implementation of sonoelastography:
for B-mode ultrasound BIRADS category 3 findings with a negative elastogram (Tsukuba
score 1 – 3), histological diagnostic confirmation (core biopsy) is not required [8].
Our own results prove that strain ratio offers lower diagnostic validity for tumors
located close to the surface of the skin. To our knowledge, these findings have not
yet been described. In a study by Chang et al, breast thickness at the location of
the target lesion was the most important factor influencing elasticity image quality.
In univariate analysis, higher lesion depth and breast thickness as well as large
tumor size were associated with low-quality elastograms, which in turn provided lower
sensitivity than high-quality elastograms [24]. In our study, we were unable to confirm the extent to which the diagnostic performance
of strain ratio is dependent on the tumor size.
Strain ratio shows significant differences in the various histological entities. A
good differentiation of malignant versus benign findings is achieved for invasive
ductal carcinomas and invasive lobular carcinomas. The latter could be a valuable
addition, as invasive lobular carcinoma is often insufficiently diagnosed when using
conventional methods (mammography, B-mode ultrasound). For mucinous carcinomas, strain
ratio outperforms 5-point score elastography. However, the results must be interpreted
with caution due to the low number of cases (n = 3). In contrast to the study by Mori
[25], our own data supports that the elastography score is not suitable for differentiating
between mucinous breast carcinomas and fibroadenomas.
In terms of false positive findings, the histological finding categories of fat necrosis
and radial scars appear to be problematic. While fat necrosis is reliably diagnosed
using mammography, radial scars remain a diagnostic challenge that, ultimately, can
only be verified as benign by means of histological methods.
There are limitations in the retrospective design of the study and in the evaluation
of the elastography data by just one investigator. It is not possible to provide data
on interobserver variability. In addition, it is difficult to compare the results
directly with other working groups as different cutoff values have been established
for the strain ratio depending on the device [7]
[16]
[17]
[18]. This means that, for the time being, strain ratio will continue to be implemented
in the diagnostic algorithm for differentiating breast lesions on a center and device-specific
basis. No general recommendation can be provided due to a lack of standardization
in the examination process, including in the software required for computation.
„Shear wave ultrasound elastography“ facilitates quantitative measurement of tissue
elasticity and represents an interesting development in this regard [23]. A recently published multi-center study by Berg et al. [26] showed that the addition of shear wave elastography to B-mode ultrasound improved
specificity without reducing sensitivity. Moreover, shear wave elastography reveals
a high intra- and interobserver reproducibility [27].
