Ultrasound in Med. & Biol., Vol. 40, No. 2, pp. 287–292, 2014 Crown Copyright Ó 2014 Published by Elsevier Inc. on behalf of World Federation for Ultrasound in Medicine & Biology Printed in the USA. All rights reserved 0301-5629/$ - see front matter
http://dx.doi.org/10.1016/j.ultrasmedbio.2013.09.020
d
Original Contribution QUANTIFICATION OF ACOUSTIC RADIATION FORCE IMPULSE IN DIFFERENTIATING BETWEEN MALIGNANT AND BENIGN BREAST LESIONS ZHENCAI LI,* JUNZHONG SUN,y JING ZHANG,z DONGMEI HU,* QIONG WANG,* and KUN PENGx * Department of Ultrasound, the First Affiliated Hospital of Chinese PLA General Hospital, Beijing, China; y Department of Oncology, the First Affiliated Hospital of Chinese PLA General Hospital, Beijing, China; z Department of Interventional Ultrasound, Chinese People’s Liberation Army General Hospital, Beijing, China; and x Department of Information, the First Affiliated Hospital of Chinese PLA General Hospital, Beijing, China (Received 12 June 2013; revised 15 September 2013; in final form 18 September 2013)
Abstract—The aim of this study was to evaluate the use of gray-level quantification (GLQ) in virtual touch tissue imaging (VTI) in the differential diagnosis of breast lesions. GLQ values of 153 lesions (101 benign, 52 malignant) were analyzed with matrix laboratory software (MATLAB, The MathWorks, Natick, MA, USA), with gray levels ranging from 0 (pure black) to 255 (pure white). The diagnostic performance of GLQ was also evaluated using receiver operating characteristic curve analysis. The mean GLQ value for benign lesions (103.27 ± 39.44) differed significantly from that for malignant lesions (44.57 ± 13.61) (p , 0.001). At a cutoff value of 52.31, the sensitivity, specificity, accuracy, positive predictive value and negative predictive value were 86.5%, 93.1%, 90.8%, 86.5% and 93.1%, respectively. In conclusion, we have proposed a method for quantification of gray levels in VTI for the differential diagnosis of breast lesions. Our results indicate that this method has the potential to aid in the classification of benign and malignant breast masses. (E-mail: [email protected]) Crown Copyright Ó 2014 Published by Elsevier Inc. on behalf of World Federation for Ultrasound in Medicine & Biology Key Words: Breast lesion, Acoustic radiation force impulse, Quantification, Gray level, Ultrasound, Strain elastography.
(acoustic radiation force impulse) (Barr et al. 2012; Jin et al. 2012). In acoustic radiation force impulse (ARFI) imaging, tissue stiffness is evaluated by using short-duration acoustic pulses to generate small (1–10 mm) localized tissue displacements instead of external compression (Nightingale et al. 2002). The tissue’s response to this radiation force is observed using conventional B-mode US imaging pulses to track tissue displacement, which correlates with local stiffness of the tissue. The waves generated can produce either a qualitative response (gray-scale map) by virtual touch tissue imaging (VTI) or a quantitative response by virtual touch tissue quantification (VTQ), depending on how they interact with the transducer (Palmeri et al. 2008; Zhai et al. 2008). In VTI, tissue stiffness is expressed using gray-level imaging, which should not be affected by equipment gain. The greater the stiffness, the darker is the gray level (Tozaki et al. 2011a, 2011b). Tissue stiffness for VTQ is expressed as shear wave velocity (m/s). Measurement of the time to peak displacement at each lateral location is used to calculate the shear wave velocity of VTQ within the tissue (D’Onofrio et al. 2010).
INTRODUCTION Ultrasonography (US) has emerged as the most important adjunct to mammography in the diagnosis of breast lesions (Flobbe et al. 2002). Two-dimensional US can be used to obtain morphologic information on the breast, but cannot determine tissue stiffness. In general, benign lesions of the breast tend to be softer than malignant lesions (Sewell 1995). Breast elastography is a new technique that can be used to determine the stiffness of targeted areas and has been found to be of clinical value in characterizing breast lesions as benign or malignant (Alhabshi et al. 2013; Barr and Zhang 2012; Barr et al. 2012; Parajuly et al. 2012). In strain elastography imaging (EI), a compressive force is applied to the tissue, and the shape-deforming effect on the tissue is measured, yielding a qualitative value for the stiffness of the lesion. The compressive force can be applied manually or with a ‘‘push’’ ultrasound force
Address correspondence to: Zhencai Li, No.51 Fucheng Road, Beijing, China 100048. E-mail: [email protected] 287
288
Ultrasound in Medicine and Biology
Because the gray level of virtual touch images is correlated with tissue stiffness, we hypothesized that gray-level quantification (GLQ) can be used to differentiate benign from malignant lesions. To the best of our knowledge, previous studies of VTI have focused mainly on area ratio (EI/B-mode) and semi-quantification of gray levels (Barr 2010; Meng et al. 2011; Tozaki et al. 2011a, 2011b). Classification of benign and malignant breast tumors using GLQ in VTI has not been reported. The main purpose of this study, therefore, was to assess the potential of GLQ in such classification. METHODS Patients The study was approved by the ethics committee of our hospital, and all the patients gave informed consent. Between December 2010 and June 2013, ARFI imaging was performed after conventional B-mode US in 151 consecutive women with breast lesions. Four patients were excluded from the study group for a lack of pathologic results. The final study group consisted of 147 patients (mean age: 45.3 y, range: 21–80 y) with a total of 153 breast lesions (101 benign, 52 malignant). Pathologic diagnoses All breast lesions were confirmed histologically by means of surgery or biopsy. Lesions were first classified as malignant or benign and then were divided into subgroups as described in Table 1. All diagnoses were made by a pathologist with 15 y of experience in breast pathologic examination. Data acquisition Virtual touch images were obtained with the Acuson S2000 US system (Siemens Medical Solutions, Mountain View, CA, USA) with a linear array transducer (9 L4, 7–9 MHz) by one sonographer with 25 y of experience in the performance of breast ultrasound. For data acquisition, conventional US scanning was perTable 1. Histologic results for malignant and benign breast lesions* Benign lesions (n 5 101)
Malignant lesions (n 5 52)
Histopathologic diagnosis
n
Histopathologic diagnosis
n
Fibroadenoma Mastitis Fibrocystic mastopathy Lobular hyperplasia
76 9 7 9
Invasive ductal carcinoma Ductal carcinoma in situ Medullary carcinoma
43 4 5
* Sensitivity 5 86.5%, specificity 5 93.1%, accuracy 5 90.8%, positive predictive value 5 86.5%, negative predictive value 5 93.1%, true positives 5 45, false positives 5 7, true negatives 5 94, false negatives 5 7.
Volume 40, Number 2, 2014
formed with patients in the supine position with both breasts fully exposed. When a clear image of the target lesion appeared on the screen, the VTI button was pressed and a region of interest (ROI) was delineated around the lesion, making sure that the ROI included sufficient surrounding breast tissue. For good imaging results, the probe should be applied with slight pressure and the patient should hold her breath; the virtual touch image appears to the right of the corresponding B-mode US image on the monitor. After each examination, the sonographer who performed elastography selected one representative image of the lesion. The image was sent to a picture archiving and communication system and saved as bit-map file on hard disk for further quantitative analysis. Computer-aided quantification Images were analyzed using matrix laboratory software (MATLAB, The MathWorks, Natick, MA, USA). The processing procedure was programmed, and the mask and histogram tools of this software were used. The details of the operating procedure are as follows: First, the virtual touch image is opened using MATLAB. Second, the cursor is placed over the area of the lesion, the enter key is pressed once and the cursor is moved continuously in the lesion area until this area is completely red. The left key of the mouse is then pressed once, and the mean gray level value of the lesion is automatically displayed on the monitor (Woo et al. 2010). Intra- and inter-observer reproducibility of method To determine intra-observer reproducibility, two GLQs were performed for each lesion, for all patients, by the same operator. Inter-observer reproducibility was also assessed independently by two operators, who performed GLQs for every lesion in all patients. Statistical analysis Statistical analyses were carried out using professional statistical software (SPSS for Microsoft Windows, Version 19.0, Chicago, IL, USA). A p-value , 0.05 was considered to indicate a statistically significant difference. Means and standard deviations of GLQ values were calculated for the malignant and benign lesions. Significant differences in values between malignant and benign lesions were evaluated with an independentsample Student t-test. Intra- and inter-observer reproducibility was assessed using correlation coefficient analysis. Receiver operating characteristic (ROC) curve analysis was adopted to evaluate the ability of the GLQ value to distinguish between malignant and benign lesions. The best cutoff point was obtained from ROC curve analysis using Youden’s index.
ARFI quantification in breast lesion differentiation d Z. LI et al.
289
RESULTS
DISCUSSION
The intra-observer difference in measuring GLQ was 6.31 6 4.06 (range: 1.56–21.38), and the interobserver difference was 8.44 6 6.05 (range: 1.87– 25.18). The corresponding correlation coefficients were 0.925 for intra-observer measurement and 0.897 for inter-observer measurement. Gray levels of malignant lesions were darker than those of benign lesions. The mean GLQ value of the malignant lesions (44.57 6 13.61) was statistically lower than that of the benign lesions (103.27 6 39.44) (p , 0.001) (Figs. 1 and 2). According to the ROC curve, the GLQ cutoff value differentiating malignant from benign lesions was estimated to be 52.31. At this cutoff value, GLQ can predict malignancy with a sensitivity of 86.5%, specificity of 93.1%, accuracy of 90.8%, positive predictive value (PPV) of 86.5% and negative predictive value (NPV) of 93.1%, yielding an area under the curve of 0.945 (Fig. 3).
Ophir et al. (1991) introduced the concept of US elastography. The traditional elastography system requires an external manual force for tissue displacement, uses different colors to indicate tissue stiffness (Garra et al. 1997) and employs a 5-point scoring system as diagnostic criteria (Zhi et al. 2010). However, both false positives and false negatives can result from differences in the external manual force applied and the physicians’ subjectivity (Barr et al. 2012; Ciurea et al. 2011; Meng et al. 2011). Instead of external manual pressure, an US equipment probe generating a short-duration acoustic radiation force to produce displacements in ROIs is used in VTI. On the basis of lesion stiffness in these areas, a corresponding gray-scale virtual touch image is generated. The greater the stiffness, the darker is the virtual touch image. In contrast, less stiff lesions result in brighter images.
Fig. 1. Image of a fibroadenoma from a 23-y-old woman. (a) Left: B-Mode ultrasonography reveals a homogeneous, hypo-echoic mass with regular margin. Right: Virtual touch tissue imaging reveals a much brighter lesion. (b) The lesion area on the virtual touch image was masked by the red color. (c) Selected virtual touch image of the lesion for which the mean gray level was 96.00 (arrow).
290
Ultrasound in Medicine and Biology
Volume 40, Number 2, 2014
Fig. 2. Image of an infiltrating carcinoma lesion from a 70-y-old woman. (a) Left: B-Mode ultrasonography depicts a heterogeneous, hypo-echoic mass with an irregular margin. Right: The virtual touch tissue image reveals a rather darker lesion. (b) The lesion area of the virtual touch image was masked by the red color. (c) Selected virtual touch image of the lesion for which the mean gray level was 29.99 (arrow).
To illustrate the value of virtual touch gray-scale images, a semi-quantitative method has been used to clas-
Fig. 3. Receiver operating characteristic (ROC) curve. The area under the curve was 0.945.
sify these images into three different patterns (Tozaki et al. 2011a, 2011b): pattern I (bright), pattern II (midbright) and pattern III (dark). The results revealed that all of the malignant lesions fall under pattern III (22/ 22), whereas most of the benign lesions fall under patterns I and II (10/18). In another article, Tozaki et al. (2012) classified virtual touch images into four patterns and obtained similar results. The authors thus concluded that the level of brightness in gray-scale virtual touch images can be used to distinguish between malignant and benign breast lesions. According to the three-pattern classification, the 52 malignant lesions in this study fall under patterns II (14/ 52) and III (38/52), whereas all 101 benign lesions fall under patterns I (44/101), II (42/101) and III (15/101), with a differential sensitivity of 73.1% and specificity of 85.1%. Although the above-mentioned semi-quantitative method is a promising ultrasound technique for differential diagnosis of breast lesions, it cannot distinguish
ARFI quantification in breast lesion differentiation d Z. LI et al.
virtual touch images with similar gray levels, which is also somewhat subjective. MATLAB software can be used to quantify the mean pixel gray level, a quantitative parameter, in a selected area (Woo et al. 2010). The gray level (brightness) of the virtual touch image determines the objective numerical value of the virtual touch image. In other words, the brighter the virtual touch image, the higher is the value; the darker the image, the lower is the value. A MATLAB histogram comprises 256 different brightness levels (gray levels) from black to white, with 0 being pure black and 255 being pure white, because standard 8-bit images contain 256 possible brightness values. That value is an objective parameter of the gray level. Thus, this objective parameter in virtual touch images can be any numerical value between 0 and 255. Histograms have been used in the quantification of echo intensity in conventional B-mode US (Chang et al. 2007; Furtado et al. 2010; Kuwata et al. 2010). The latter studies reported that stronger echo intensities are associated with higher gray-level numerical values, whereas weaker echo intensities are associated with lower gray-level numerical values. In this study, the virtual touch images of malignant lesions were darker than those of benign lesions. The mean values for benign and malignant lesions were 103.27 6 39.44 and 44.57 6 13.61, respectively; the difference between these values is significant (p , 0.001). At a GLQ cutoff value of 52.31, this approach predicted malignancy with a sensitivity of 86.5%, specificity of 93.1%, accuracy of 90.8%, PPV of 86.5% and NPV of 93.1%, and yielded an area under the curve of 0.945. These results indicate that GLQ has a higher sensitivity and specificity than the semi-quantitative method in the differential diagnosis of malignant masses in the breast. The intra-observer (0.925) and inter-observer (0.897) correlation coefficients confirmed the reproducibility of this quantitative method. Some limitations of the present study should be considered. First, for GLQ, the exact boundaries of lesions must be identified, as errors may occur when the boundaries are vague. In this case, the conventional B-mode US image on the left of the display is important in recognizing the boundaries of the lesion. Second, the virtual touch image was quantified as the mean pixel gray level in the lesion area. Calcification and necrosis in the lesions may result in errors. Finally, this study was limited by the relatively small number of cases. Further studies on a larger scale are needed to assess fully the role of GLQ in the differential diagnosis of breast lesions. CONCLUSIONS We have proposed a method for quantification of virtual touch images for the differential diagnosis of breast
291
lesions. Our results indicate that this method has the potential to aid in the classification of benign and malignant breast masses.
REFERENCES Alhabshi SM, Rahmat K, Abdul Halim N, Aziz S, Radhika S, Gan GC, Vijayananthan A, Westerhout CJ, Mohd-Shah MN, Jaszle S, Harlina Mohd Latar N, Muhammad R. Semi-quantitative and qualitative assessment of breast ultrasound elastography in differentiating between malignant and benign lesions. Ultrasound Med Biol 2013; 39:568–578. Barr RG. Real-time ultrasound elasticity of the breast-initial clinical results. Ultrasound Q 2010;26:61–66. Barr RG, Destounis S, Lackey LB II, Svensson WE, Balleyguier C, Smith C. Evaluation of breast lesions using sonographic elasticity imaging. J Ultrasound Med 2012;31:281–287. Barr RG, Zhang Z. Effects of precompression on elasticity imaging of the breast. J Ultrasound Med 2012;31:895–902. Chang RF, Huang SF, Moon WK, Lee YH, Chen DR. Solid breast masses: Neural network analysis of vascular features at threedimensional power Doppler US for benign or malignant classification. Radiology 2007;243:56–62. Ciurea AL, Bolboaca SD, Ciortea CA, Botar-Jid C, Dudea SM. The influence of technical factors on sonoelastographic assessment of solid breast nodules. Ultraschall Med 2011;32(Suppl 1):S27–S34. D’Onofrio M, Gallotti A, Pozzi Mucelli R. Virtual touch tissue quantification: Measurement repeatability and normal values in the healthy liver. AJR Am J Roentgenol 2010;195:132–136. Flobbe K, Nelemans PJ, Kessels AG, Beets GL, Von Meyenfeldt MF, Van Engelshoven JM. The role of ultrasonography as an adjunct to mammography in the detection of breast cancer: A systematic review. Eur J Cancer 2002;38:1044–1050. Furtado MR, Pires CR, Araujo Junior E, De Souza E, Nardozza LM, Moron AF. Transvaginal grey scale histogram of the cervix at 2025 weeks of pregnancy. Aust NZ J Obstet Gynaecol 2010;50: 444–449. Garra B, Cespedes E, Ophir J, Spratt S, Zuurbier R, Magnant C, Pennanen M. Elastography of breast lesions: Initial clinical results. Radiology 1997;202:79–86. Jin ZQ, Li XR, Zhou HL, Chen JX, Huang X, Dai HX, Li JW, Chen XD, Xu XH. Acoustic radiation force impulse elastography of breast imaging reporting and data system category 4 breast lesions. Clin Breast Cancer 2012;12:420–427. Kuwata T, Matsubara S, Taniguchi N, Ohkuchi A, Ohkusa T, Suzuki M. A novel method for evaluating uterine cervical consistency using vaginal ultrasound gray-level histogram. J Perinat Med 2010;38: 491–494. Meng W, Zhang G, Wu C, Wu G, Song Y, Lu Z. Preliminary results of acoustic radiation force impulse (ARFI) ultrasound imaging of breast lesions. Ultrasound Med Biol 2011;37:1436–1443. Nightingale K, Soo MS, Nightingale R, Trahey G. Acoustic radiation force impulse imaging: in vivo demonstration of clinical feasibility. Ultrasound Med Biol 2002;28:227–235. Ophir J, Cespedes I, Ponnekanti H, Yazdi Y, Li X. Elastography: A quantitative method for imaging the elasticity of biological tissues. Ultrason Imaging 1991;13:111–134. Palmeri ML, Wang MH, Dahl JJ, Frinkley KD, Nightingale KR. Quantifying hepatic shear modulus in vivo using acoustic radiation force. Ultrasound Med Biol 2008;34:546–558. Parajuly SS, Lan PY, Yun MB, Gang YZ, Hha Z. Diagnostic potential of strain ratio measurement and 5 point scoring method for detection of breast cancer: Chinese experience. Asian Pac J Cancer Prev 2012; 13:1447–1452. Sewell CW. Pathology of benign and malignant breast disorders. Radiol Clin North Am 1995;33:1067–1080. Tozaki M, Isobe S, Fukuma E. Preliminary study of ultrasonographic tissue quantification of the breast using the acoustic
292
Ultrasound in Medicine and Biology
radiation force impulse (ARFI) technology. Eur J Radiol 2011a; 80:182–187. Tozaki M, Isobe S, Yamaguchi M, Ogawa Y, Homma K, Saito M, Joo C, Fukuma E. Ultrasonographic elastography of the breast using acoustic radiation force impulse technology: Preliminary study. Jpn J Radiol 2011b;29:452–456. Tozaki M, Isobe S, Sakamoto M. Combination of elastography and tissue quantification using the acoustic radiation force impulse (ARFI) technology for differential diagnosis of breast masses. Jpn J Radio 2012;30:659–670.
Volume 40, Number 2, 2014 Woo KM, Ji WC, Nariya C, Sang HP, Jung MC, Mijung J, Kwang GK. Computer-aided analysis of ultrasound elasticity images for classification of benign and malignant breast masses. AJR Am J Roentgenol 2010;195:1460–1465. Zhai L, Palmeri ML, Bouchard RR, Nightingale RW, Nightingale KR. An integrated indenter-ARFI imaging system for tissue stiffness quantification. Ultrason Imaging 2008;30:95–111. Zhi H, Xiao XY, Yang HY, Ou B, Wen YL, Luo BM. Ultrasonic elastography in breast cancer diagnosis: Strain ratio vs 5-point scale. Acad Radiol 2010;17:1227–1233.
http://dx.doi.org/10.1016/j.ultrasmedbio.2013.09.020
d
Original Contribution QUANTIFICATION OF ACOUSTIC RADIATION FORCE IMPULSE IN DIFFERENTIATING BETWEEN MALIGNANT AND BENIGN BREAST LESIONS ZHENCAI LI,* JUNZHONG SUN,y JING ZHANG,z DONGMEI HU,* QIONG WANG,* and KUN PENGx * Department of Ultrasound, the First Affiliated Hospital of Chinese PLA General Hospital, Beijing, China; y Department of Oncology, the First Affiliated Hospital of Chinese PLA General Hospital, Beijing, China; z Department of Interventional Ultrasound, Chinese People’s Liberation Army General Hospital, Beijing, China; and x Department of Information, the First Affiliated Hospital of Chinese PLA General Hospital, Beijing, China (Received 12 June 2013; revised 15 September 2013; in final form 18 September 2013)
Abstract—The aim of this study was to evaluate the use of gray-level quantification (GLQ) in virtual touch tissue imaging (VTI) in the differential diagnosis of breast lesions. GLQ values of 153 lesions (101 benign, 52 malignant) were analyzed with matrix laboratory software (MATLAB, The MathWorks, Natick, MA, USA), with gray levels ranging from 0 (pure black) to 255 (pure white). The diagnostic performance of GLQ was also evaluated using receiver operating characteristic curve analysis. The mean GLQ value for benign lesions (103.27 ± 39.44) differed significantly from that for malignant lesions (44.57 ± 13.61) (p , 0.001). At a cutoff value of 52.31, the sensitivity, specificity, accuracy, positive predictive value and negative predictive value were 86.5%, 93.1%, 90.8%, 86.5% and 93.1%, respectively. In conclusion, we have proposed a method for quantification of gray levels in VTI for the differential diagnosis of breast lesions. Our results indicate that this method has the potential to aid in the classification of benign and malignant breast masses. (E-mail: [email protected]) Crown Copyright Ó 2014 Published by Elsevier Inc. on behalf of World Federation for Ultrasound in Medicine & Biology Key Words: Breast lesion, Acoustic radiation force impulse, Quantification, Gray level, Ultrasound, Strain elastography.
(acoustic radiation force impulse) (Barr et al. 2012; Jin et al. 2012). In acoustic radiation force impulse (ARFI) imaging, tissue stiffness is evaluated by using short-duration acoustic pulses to generate small (1–10 mm) localized tissue displacements instead of external compression (Nightingale et al. 2002). The tissue’s response to this radiation force is observed using conventional B-mode US imaging pulses to track tissue displacement, which correlates with local stiffness of the tissue. The waves generated can produce either a qualitative response (gray-scale map) by virtual touch tissue imaging (VTI) or a quantitative response by virtual touch tissue quantification (VTQ), depending on how they interact with the transducer (Palmeri et al. 2008; Zhai et al. 2008). In VTI, tissue stiffness is expressed using gray-level imaging, which should not be affected by equipment gain. The greater the stiffness, the darker is the gray level (Tozaki et al. 2011a, 2011b). Tissue stiffness for VTQ is expressed as shear wave velocity (m/s). Measurement of the time to peak displacement at each lateral location is used to calculate the shear wave velocity of VTQ within the tissue (D’Onofrio et al. 2010).
INTRODUCTION Ultrasonography (US) has emerged as the most important adjunct to mammography in the diagnosis of breast lesions (Flobbe et al. 2002). Two-dimensional US can be used to obtain morphologic information on the breast, but cannot determine tissue stiffness. In general, benign lesions of the breast tend to be softer than malignant lesions (Sewell 1995). Breast elastography is a new technique that can be used to determine the stiffness of targeted areas and has been found to be of clinical value in characterizing breast lesions as benign or malignant (Alhabshi et al. 2013; Barr and Zhang 2012; Barr et al. 2012; Parajuly et al. 2012). In strain elastography imaging (EI), a compressive force is applied to the tissue, and the shape-deforming effect on the tissue is measured, yielding a qualitative value for the stiffness of the lesion. The compressive force can be applied manually or with a ‘‘push’’ ultrasound force
Address correspondence to: Zhencai Li, No.51 Fucheng Road, Beijing, China 100048. E-mail: [email protected] 287
288
Ultrasound in Medicine and Biology
Because the gray level of virtual touch images is correlated with tissue stiffness, we hypothesized that gray-level quantification (GLQ) can be used to differentiate benign from malignant lesions. To the best of our knowledge, previous studies of VTI have focused mainly on area ratio (EI/B-mode) and semi-quantification of gray levels (Barr 2010; Meng et al. 2011; Tozaki et al. 2011a, 2011b). Classification of benign and malignant breast tumors using GLQ in VTI has not been reported. The main purpose of this study, therefore, was to assess the potential of GLQ in such classification. METHODS Patients The study was approved by the ethics committee of our hospital, and all the patients gave informed consent. Between December 2010 and June 2013, ARFI imaging was performed after conventional B-mode US in 151 consecutive women with breast lesions. Four patients were excluded from the study group for a lack of pathologic results. The final study group consisted of 147 patients (mean age: 45.3 y, range: 21–80 y) with a total of 153 breast lesions (101 benign, 52 malignant). Pathologic diagnoses All breast lesions were confirmed histologically by means of surgery or biopsy. Lesions were first classified as malignant or benign and then were divided into subgroups as described in Table 1. All diagnoses were made by a pathologist with 15 y of experience in breast pathologic examination. Data acquisition Virtual touch images were obtained with the Acuson S2000 US system (Siemens Medical Solutions, Mountain View, CA, USA) with a linear array transducer (9 L4, 7–9 MHz) by one sonographer with 25 y of experience in the performance of breast ultrasound. For data acquisition, conventional US scanning was perTable 1. Histologic results for malignant and benign breast lesions* Benign lesions (n 5 101)
Malignant lesions (n 5 52)
Histopathologic diagnosis
n
Histopathologic diagnosis
n
Fibroadenoma Mastitis Fibrocystic mastopathy Lobular hyperplasia
76 9 7 9
Invasive ductal carcinoma Ductal carcinoma in situ Medullary carcinoma
43 4 5
* Sensitivity 5 86.5%, specificity 5 93.1%, accuracy 5 90.8%, positive predictive value 5 86.5%, negative predictive value 5 93.1%, true positives 5 45, false positives 5 7, true negatives 5 94, false negatives 5 7.
Volume 40, Number 2, 2014
formed with patients in the supine position with both breasts fully exposed. When a clear image of the target lesion appeared on the screen, the VTI button was pressed and a region of interest (ROI) was delineated around the lesion, making sure that the ROI included sufficient surrounding breast tissue. For good imaging results, the probe should be applied with slight pressure and the patient should hold her breath; the virtual touch image appears to the right of the corresponding B-mode US image on the monitor. After each examination, the sonographer who performed elastography selected one representative image of the lesion. The image was sent to a picture archiving and communication system and saved as bit-map file on hard disk for further quantitative analysis. Computer-aided quantification Images were analyzed using matrix laboratory software (MATLAB, The MathWorks, Natick, MA, USA). The processing procedure was programmed, and the mask and histogram tools of this software were used. The details of the operating procedure are as follows: First, the virtual touch image is opened using MATLAB. Second, the cursor is placed over the area of the lesion, the enter key is pressed once and the cursor is moved continuously in the lesion area until this area is completely red. The left key of the mouse is then pressed once, and the mean gray level value of the lesion is automatically displayed on the monitor (Woo et al. 2010). Intra- and inter-observer reproducibility of method To determine intra-observer reproducibility, two GLQs were performed for each lesion, for all patients, by the same operator. Inter-observer reproducibility was also assessed independently by two operators, who performed GLQs for every lesion in all patients. Statistical analysis Statistical analyses were carried out using professional statistical software (SPSS for Microsoft Windows, Version 19.0, Chicago, IL, USA). A p-value , 0.05 was considered to indicate a statistically significant difference. Means and standard deviations of GLQ values were calculated for the malignant and benign lesions. Significant differences in values between malignant and benign lesions were evaluated with an independentsample Student t-test. Intra- and inter-observer reproducibility was assessed using correlation coefficient analysis. Receiver operating characteristic (ROC) curve analysis was adopted to evaluate the ability of the GLQ value to distinguish between malignant and benign lesions. The best cutoff point was obtained from ROC curve analysis using Youden’s index.
ARFI quantification in breast lesion differentiation d Z. LI et al.
289
RESULTS
DISCUSSION
The intra-observer difference in measuring GLQ was 6.31 6 4.06 (range: 1.56–21.38), and the interobserver difference was 8.44 6 6.05 (range: 1.87– 25.18). The corresponding correlation coefficients were 0.925 for intra-observer measurement and 0.897 for inter-observer measurement. Gray levels of malignant lesions were darker than those of benign lesions. The mean GLQ value of the malignant lesions (44.57 6 13.61) was statistically lower than that of the benign lesions (103.27 6 39.44) (p , 0.001) (Figs. 1 and 2). According to the ROC curve, the GLQ cutoff value differentiating malignant from benign lesions was estimated to be 52.31. At this cutoff value, GLQ can predict malignancy with a sensitivity of 86.5%, specificity of 93.1%, accuracy of 90.8%, positive predictive value (PPV) of 86.5% and negative predictive value (NPV) of 93.1%, yielding an area under the curve of 0.945 (Fig. 3).
Ophir et al. (1991) introduced the concept of US elastography. The traditional elastography system requires an external manual force for tissue displacement, uses different colors to indicate tissue stiffness (Garra et al. 1997) and employs a 5-point scoring system as diagnostic criteria (Zhi et al. 2010). However, both false positives and false negatives can result from differences in the external manual force applied and the physicians’ subjectivity (Barr et al. 2012; Ciurea et al. 2011; Meng et al. 2011). Instead of external manual pressure, an US equipment probe generating a short-duration acoustic radiation force to produce displacements in ROIs is used in VTI. On the basis of lesion stiffness in these areas, a corresponding gray-scale virtual touch image is generated. The greater the stiffness, the darker is the virtual touch image. In contrast, less stiff lesions result in brighter images.
Fig. 1. Image of a fibroadenoma from a 23-y-old woman. (a) Left: B-Mode ultrasonography reveals a homogeneous, hypo-echoic mass with regular margin. Right: Virtual touch tissue imaging reveals a much brighter lesion. (b) The lesion area on the virtual touch image was masked by the red color. (c) Selected virtual touch image of the lesion for which the mean gray level was 96.00 (arrow).
290
Ultrasound in Medicine and Biology
Volume 40, Number 2, 2014
Fig. 2. Image of an infiltrating carcinoma lesion from a 70-y-old woman. (a) Left: B-Mode ultrasonography depicts a heterogeneous, hypo-echoic mass with an irregular margin. Right: The virtual touch tissue image reveals a rather darker lesion. (b) The lesion area of the virtual touch image was masked by the red color. (c) Selected virtual touch image of the lesion for which the mean gray level was 29.99 (arrow).
To illustrate the value of virtual touch gray-scale images, a semi-quantitative method has been used to clas-
Fig. 3. Receiver operating characteristic (ROC) curve. The area under the curve was 0.945.
sify these images into three different patterns (Tozaki et al. 2011a, 2011b): pattern I (bright), pattern II (midbright) and pattern III (dark). The results revealed that all of the malignant lesions fall under pattern III (22/ 22), whereas most of the benign lesions fall under patterns I and II (10/18). In another article, Tozaki et al. (2012) classified virtual touch images into four patterns and obtained similar results. The authors thus concluded that the level of brightness in gray-scale virtual touch images can be used to distinguish between malignant and benign breast lesions. According to the three-pattern classification, the 52 malignant lesions in this study fall under patterns II (14/ 52) and III (38/52), whereas all 101 benign lesions fall under patterns I (44/101), II (42/101) and III (15/101), with a differential sensitivity of 73.1% and specificity of 85.1%. Although the above-mentioned semi-quantitative method is a promising ultrasound technique for differential diagnosis of breast lesions, it cannot distinguish
ARFI quantification in breast lesion differentiation d Z. LI et al.
virtual touch images with similar gray levels, which is also somewhat subjective. MATLAB software can be used to quantify the mean pixel gray level, a quantitative parameter, in a selected area (Woo et al. 2010). The gray level (brightness) of the virtual touch image determines the objective numerical value of the virtual touch image. In other words, the brighter the virtual touch image, the higher is the value; the darker the image, the lower is the value. A MATLAB histogram comprises 256 different brightness levels (gray levels) from black to white, with 0 being pure black and 255 being pure white, because standard 8-bit images contain 256 possible brightness values. That value is an objective parameter of the gray level. Thus, this objective parameter in virtual touch images can be any numerical value between 0 and 255. Histograms have been used in the quantification of echo intensity in conventional B-mode US (Chang et al. 2007; Furtado et al. 2010; Kuwata et al. 2010). The latter studies reported that stronger echo intensities are associated with higher gray-level numerical values, whereas weaker echo intensities are associated with lower gray-level numerical values. In this study, the virtual touch images of malignant lesions were darker than those of benign lesions. The mean values for benign and malignant lesions were 103.27 6 39.44 and 44.57 6 13.61, respectively; the difference between these values is significant (p , 0.001). At a GLQ cutoff value of 52.31, this approach predicted malignancy with a sensitivity of 86.5%, specificity of 93.1%, accuracy of 90.8%, PPV of 86.5% and NPV of 93.1%, and yielded an area under the curve of 0.945. These results indicate that GLQ has a higher sensitivity and specificity than the semi-quantitative method in the differential diagnosis of malignant masses in the breast. The intra-observer (0.925) and inter-observer (0.897) correlation coefficients confirmed the reproducibility of this quantitative method. Some limitations of the present study should be considered. First, for GLQ, the exact boundaries of lesions must be identified, as errors may occur when the boundaries are vague. In this case, the conventional B-mode US image on the left of the display is important in recognizing the boundaries of the lesion. Second, the virtual touch image was quantified as the mean pixel gray level in the lesion area. Calcification and necrosis in the lesions may result in errors. Finally, this study was limited by the relatively small number of cases. Further studies on a larger scale are needed to assess fully the role of GLQ in the differential diagnosis of breast lesions. CONCLUSIONS We have proposed a method for quantification of virtual touch images for the differential diagnosis of breast
291
lesions. Our results indicate that this method has the potential to aid in the classification of benign and malignant breast masses.
REFERENCES Alhabshi SM, Rahmat K, Abdul Halim N, Aziz S, Radhika S, Gan GC, Vijayananthan A, Westerhout CJ, Mohd-Shah MN, Jaszle S, Harlina Mohd Latar N, Muhammad R. Semi-quantitative and qualitative assessment of breast ultrasound elastography in differentiating between malignant and benign lesions. Ultrasound Med Biol 2013; 39:568–578. Barr RG. Real-time ultrasound elasticity of the breast-initial clinical results. Ultrasound Q 2010;26:61–66. Barr RG, Destounis S, Lackey LB II, Svensson WE, Balleyguier C, Smith C. Evaluation of breast lesions using sonographic elasticity imaging. J Ultrasound Med 2012;31:281–287. Barr RG, Zhang Z. Effects of precompression on elasticity imaging of the breast. J Ultrasound Med 2012;31:895–902. Chang RF, Huang SF, Moon WK, Lee YH, Chen DR. Solid breast masses: Neural network analysis of vascular features at threedimensional power Doppler US for benign or malignant classification. Radiology 2007;243:56–62. Ciurea AL, Bolboaca SD, Ciortea CA, Botar-Jid C, Dudea SM. The influence of technical factors on sonoelastographic assessment of solid breast nodules. Ultraschall Med 2011;32(Suppl 1):S27–S34. D’Onofrio M, Gallotti A, Pozzi Mucelli R. Virtual touch tissue quantification: Measurement repeatability and normal values in the healthy liver. AJR Am J Roentgenol 2010;195:132–136. Flobbe K, Nelemans PJ, Kessels AG, Beets GL, Von Meyenfeldt MF, Van Engelshoven JM. The role of ultrasonography as an adjunct to mammography in the detection of breast cancer: A systematic review. Eur J Cancer 2002;38:1044–1050. Furtado MR, Pires CR, Araujo Junior E, De Souza E, Nardozza LM, Moron AF. Transvaginal grey scale histogram of the cervix at 2025 weeks of pregnancy. Aust NZ J Obstet Gynaecol 2010;50: 444–449. Garra B, Cespedes E, Ophir J, Spratt S, Zuurbier R, Magnant C, Pennanen M. Elastography of breast lesions: Initial clinical results. Radiology 1997;202:79–86. Jin ZQ, Li XR, Zhou HL, Chen JX, Huang X, Dai HX, Li JW, Chen XD, Xu XH. Acoustic radiation force impulse elastography of breast imaging reporting and data system category 4 breast lesions. Clin Breast Cancer 2012;12:420–427. Kuwata T, Matsubara S, Taniguchi N, Ohkuchi A, Ohkusa T, Suzuki M. A novel method for evaluating uterine cervical consistency using vaginal ultrasound gray-level histogram. J Perinat Med 2010;38: 491–494. Meng W, Zhang G, Wu C, Wu G, Song Y, Lu Z. Preliminary results of acoustic radiation force impulse (ARFI) ultrasound imaging of breast lesions. Ultrasound Med Biol 2011;37:1436–1443. Nightingale K, Soo MS, Nightingale R, Trahey G. Acoustic radiation force impulse imaging: in vivo demonstration of clinical feasibility. Ultrasound Med Biol 2002;28:227–235. Ophir J, Cespedes I, Ponnekanti H, Yazdi Y, Li X. Elastography: A quantitative method for imaging the elasticity of biological tissues. Ultrason Imaging 1991;13:111–134. Palmeri ML, Wang MH, Dahl JJ, Frinkley KD, Nightingale KR. Quantifying hepatic shear modulus in vivo using acoustic radiation force. Ultrasound Med Biol 2008;34:546–558. Parajuly SS, Lan PY, Yun MB, Gang YZ, Hha Z. Diagnostic potential of strain ratio measurement and 5 point scoring method for detection of breast cancer: Chinese experience. Asian Pac J Cancer Prev 2012; 13:1447–1452. Sewell CW. Pathology of benign and malignant breast disorders. Radiol Clin North Am 1995;33:1067–1080. Tozaki M, Isobe S, Fukuma E. Preliminary study of ultrasonographic tissue quantification of the breast using the acoustic
292
Ultrasound in Medicine and Biology
radiation force impulse (ARFI) technology. Eur J Radiol 2011a; 80:182–187. Tozaki M, Isobe S, Yamaguchi M, Ogawa Y, Homma K, Saito M, Joo C, Fukuma E. Ultrasonographic elastography of the breast using acoustic radiation force impulse technology: Preliminary study. Jpn J Radiol 2011b;29:452–456. Tozaki M, Isobe S, Sakamoto M. Combination of elastography and tissue quantification using the acoustic radiation force impulse (ARFI) technology for differential diagnosis of breast masses. Jpn J Radio 2012;30:659–670.
Volume 40, Number 2, 2014 Woo KM, Ji WC, Nariya C, Sang HP, Jung MC, Mijung J, Kwang GK. Computer-aided analysis of ultrasound elasticity images for classification of benign and malignant breast masses. AJR Am J Roentgenol 2010;195:1460–1465. Zhai L, Palmeri ML, Bouchard RR, Nightingale RW, Nightingale KR. An integrated indenter-ARFI imaging system for tissue stiffness quantification. Ultrason Imaging 2008;30:95–111. Zhi H, Xiao XY, Yang HY, Ou B, Wen YL, Luo BM. Ultrasonic elastography in breast cancer diagnosis: Strain ratio vs 5-point scale. Acad Radiol 2010;17:1227–1233.
