- Email: [email protected]
Breast Density Estimation with Fully Automated Volumetric Method:
ARTICLE IN PRESS
Original Investigation
Breast Density Estimation with Fully Automated Volumetric Method: Comparison to Radiologists’ Assessment by BI-RADS Categories Tulika Singh, MD, Madhurima Sharma, MD, Veenu Singla, MD, Niranjan Khandelwal, MD, DNB Rationale and Objectives: The objective of our study was to calculate mammographic breast density with a fully automated volumetric breast density measurement method and to compare it to breast imaging reporting and data system (BI-RADS) breast density categories assigned by two radiologists. Materials and Methods: A total of 476 full-field digital mammography examinations with standard mediolateral oblique and craniocaudal views were evaluated by two blinded radiologists and BI-RADS density categories were assigned. Using a fully automated software, mean fibroglandular tissue volume, mean breast volume, and mean volumetric breast density were calculated. Based on percentage volumetric breast density, a volumetric density grade was assigned from 1 to 4. Results: The weighted overall kappa was 0.895 (almost perfect agreement) for the two radiologists’ BI-RADS density estimates. A statistically significant difference was seen in mean volumetric breast density among the BI-RADS density categories. With increased BIRADS density category, increase in mean volumetric breast density was also seen (P < 0.001). A significant positive correlation was found between BI-RADS categories and volumetric density grading by fully automated software (ρ = 0.728, P < 0.001 for first radiologist and ρ = 0.725, P < 0.001 for second radiologist). Pairwise estimates of the weighted kappa between Volpara density grade and BI-RADS density category by two observers showed fair agreement (κ = 0.398 and 0.388, respectively). Conclusions: In our study, a good correlation was seen between density grading using fully automated volumetric method and density grading using BI-RADS density categories assigned by the two radiologists. Thus, the fully automated volumetric method may be used to quantify breast density on routine mammography. Key Words: Automated; BI-RADS; Breast density; Mammography; Volumetric method. © 2015 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.
INTRODUCTION
M
ammography is the recommended screening method for breast cancer in the general population (1). However, the sensitivity of screening mammography is reduced in women with dense breasts because of obscuration of underlying lesion by dense and heterogenous breast tissue. Because of this, chances of missing breast cancer are increased (2–4). At the same time, increased breast density has been considered as an independent risk factor for the development of breast cancer. Various studies have shown that women with high breast density compared to women with low breast density are four to six times more likely to get breast cancer (5–7). Bertrand et al. reported that women with breast Acad Radiol 2015; ■:■■–■■ From the Department of Radiodiagnosis and Imaging, Post Graduate Institute of Medical Education and Research (PGIMER), Chandigarh, India. Received July 3, 2015; revised September 17, 2015; accepted September 20, 2015. Address for correspondence to: T.S. e-mail: [email protected] © 2015 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.acra.2015.09.012
density more than 50% have approximately twofold risk of developing breast cancer than women with breast density of 11–23%. The authors also reported a positive correlation between breast density and larger tumor size, positive lymph nodes, and negative estrogen receptor status (8). The American College of Radiology (ACR) recommends annual screening mammography in women more than 40 years of age. However, in a subset of population having increased risk for breast cancer, ACR not only recommends screening at earlier age but also emphasizes the need of supplemental screening methods like ultrasound and magnetic resonance imaging (1). ACR has also recommended supplemental screening in women with genetic predisposition to the disease and in women with dense breasts (1). Thus, breast density determines the need for supplemental screening and can prove to be an important factor in risk stratification. Various qualitative and quantitative methods have been devised for measurement of breast density on mammography. Breast imaging reporting and data system (BI-RADS) classification system is the commonest method in practice for measurement of breast density (9). The system is qualitative 1
SINGH ET AL
and based on visual analysis. Increased score implies increased breast density. However, this method is not reliably reproducible. Various studies done in the past have shown variable interobserver agreement among the radiologists (κ = 0.43–0.76) (10–12). In one large cohort study, wide variability was found among community radiologists in measurement of breast density and only 18% had moderate to substantial agreement (12). To overcome this problem, various quantitative methods for breast density measurement have been developed: visual computer assisted (semiautomated) or fully automated. These methods can be area based or volume based. Area-based method estimates breast density from area segmentation of two-dimensional image of compressed breast (13,14). The semiautomated method is based upon user-defined threshold thereby increasing subjectivity of the method and the results are highly user dependent, with substantial inter- and intraobserver variability (11,15). In addition, both qualitative and area-based methods have certain other limitations, including two-dimensional measurements, variability with breast compression, and no consideration of breast thickness (16). Because of these limitations, various volumetric methods have been developed for measuring breast density (17,18). Fully automated volumetric methods for density measurement are based upon physical model that assumes that the breast is composed of fibroglandular parenchyma and fat. Tissue composition in a given pixel can be calculated from x-ray attenuation properties of these tissues, and fibroglandular density can thus be estimated (18,19). Volumetric method takes breast thickness into account and is more reproducible. Limited number of studies has been carried out in the past to compare various methods of breast density estimation. The purpose of our study was to compare qualitative method of breast density grading using BI-RADS to breast density estimates from fully automated volumetric software. Moreover, most of the previous studies have been done on a Korean population. The objective of our study was to validate the results of fully automated volumetric method on an Indian cohort.
MATERIALS AND METHODS Patients and Mammography
This was a retrospective study approved by our institutional ethical committee. Informed written consent was obtained from all the patients. From March 2013 to April 2014, a total of 715 full-field digital mammography (FFDM) examinations with standard views (mediolateral oblique and craniocaudal) were performed. All examinations performed on asymptomatic females more than 35 years of age were included in this study. Patients with age less than 35 years, with breast symptoms, and with previous history of any breast surgery were excluded from the study. A total of 239 patients were excluded from the study. All digital mammographic examinations were performed on an FFDM unit: MicroDose SI, Philips (Amsterdam, Netherlands). The system employs crystalline silicon 2
Academic Radiology, Vol ■, No , ■■ 2015
photon counting chamber type of detector and has spatial resolution of 50 μm and image data matrix size of 4800 × 5200 pixels. BI-RADS Density Classification The mammography images were analyzed independently by two blinded radiologists who specialize in breast imaging and had 5–10 years of experience in interpreting mammography. All the images were reviewed in Digital imaging and communication (DICOM) in medicine format at the review workstation. The images were read as a batch at the end of the study. The radiologists assessed breast density in each mammogram according to the BI-RADS breast density categories. Mediolateral oblique and craniocaudal views were read simultaneously to assess the breast density. The following BIRADS categories for breast density were used for mammographic interpretations: category 1, almost fatty; category 2, scattered fibroglandular densities; category 3, heterogeneously dense; and category 4, extremely dense (20). Mammographic Density Analysis by Fully Automated Volumetric Software
For fully automated volumetric analysis, Volpara software (version 1.4.5, Ma¯takina Technology, LTD, Wellington, New Zealand) was used. The Volpara software processes the image data generated by the digital mammography system and calculates the breast density. The measurement starts with finding a reference point in the breast having known composition (usually near the chest wall containing all fatty tissue). Then x-ray attenuation is calculated in each pixel. From degree of attenuation in a pixel, composition of tissue located between the pixel and the x-ray source is estimated and density maps are created. By analyzing values in density maps, the software computes the fibroglandular tissue volume, the breast volume (both in cubic centimetres), and the volumetric breast density. The volumetric density is then computed from these data, which range from 0% to 40%. Mediolateral oblique and craniocaudal views are averaged and the density information is provided per breast. A Volpara density grade (VDG) is also given to each patient. Total fibroglandular tissue volume is divided by total breast volume to obtain a percentage per patient. The resultant percentage is graded as VDG as follows: 0–4.7% volumetric density, VDG 1; 4.8–7.9%, VDG 2; 8.0– 15.0%, VDG 3; and 15.1% and above, VDG 4. After completion of the study, Volpara automatically processes data and sends DICOM secondary capture images containing the breast density information (Fig. 1). Data and Statistical Analysis
Interobserver agreement in measurement of breast density using BI-RADS was calculated using weighted kappa. The kappa values were interpreted as suggested by Landis and Koch (21) as follows: a kappa value equal to or less than 0.20 indicates
Academic Radiology, Vol ■, No , ■■ 2015
BREAST DENSITY ESTIMATION
TABLE 1. Table Showing Age of the Patients and Various Parameters Obtained by Automated Breast Density Calculation Using Volpara Software (Version 1.4.5) Age (mean ± SD) Volume of fibroglandular tissue (mean ± SD) Breast volume (mean ± SD) Volumetric breast density (mean ± SD)
48.8 ± 7.07 3.999 ± 2.529 57.905 ± 25.325 7.676 ± 4.891
Figure 1. Image showing example of breast density estimation using fully automated volumetric method (Volpara version 1.4.5).
Figure 3. 1–4.
Figure 2. Age distribution of patients included in the study.
slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, almost perfect agreement. Statistical comparisons were performed with independent and paired Student t tests for continuous variables and the chi-square or Fisher exact test for categorical variables. The correlation between the BIRADS density category and the volumetric breast density provided by the fully automated software was estimated using the Spearman rank correlation coefficient (ρ). Statistical analysis was performed using SPSS Statistics 17. Differences were considered to be statistically significant at P < 0.05. RESULTS A total of 476 patients were enrolled in the study. The age range of patients was from 36 to 76 years with a mean of 48.8 years and a standard deviation of 7.07 years (Fig. 2). Most of the patients were in the age range of 41–50 years (53.4%) and 51–60 years (29.6%). Using fully automated software (Volpara, version 1.4.5), mean volume of fibroglandular tissue, breast volume, and volumetric breast density were derived and have been summarized
Bar chart showing frequency of patients in VDG grade
in Table 1. According to volumetric density, VDG grade from 1 to 4 was assigned (Fig. 3). Interobserver agreement between two radiologists for the assignment of BI-RADS categories was almost perfect (κ = 0.895). Pairwise estimates of the weighted kappa between VDG grade and BI-RADS density category by two observers showed fair agreement (κ = 0.398 and 0.388, respectively). On visual assessment, less than one fourth of the study population was categorized as BI-RADS 3 or BI-RADS 4 (Table 2). According to the results from the first radiologist’s grading of BI-RADS density category, BI-RADS category 3 was found in 20.6% of the cases and BI-RADS category 4 was found in 1.5% of the cases. Similarly, according to the second radiologist, 18.9% of the patients were assigned BI-RADS category 3 and 0.6% of the patients were assigned BIRADS category 4 (Table 2). However, automated volumetric assessment from the Volpara software showed ~41% of the patients having VDG grade 3 (34.5%) and grade 4 (6.9%) (Fig. 3). Out of 476 examinations, 444 examinations (93.3%) showed agreement between BI-RADS grading of both the radiologists. In the remaining 32 patients with disagreement, differences were within one category only. Mean volumetric breast density of all patients in each BI-RADS category was also calculated (by agreement of both radiologists) (Table 3). There was a significant difference between volumetric breast density of 3
Academic Radiology, Vol ■, No , ■■ 2015
SINGH ET AL
TABLE 2. Table Showing Number of Patients in Each Grade According to Volpara Density Grade (VDG) and BI-RADS Grading by Two Observers
VDG Grade
Number of Patients (Percentage)
BI-RADS Grade
Observer 1
Observer 2
148 (31.1) 131 (27.5) 164 (34.5) 33 (6.9)
1 2 3 4
159 (33.4) 212 (44.5) 98 (20.6) 7 (1.5)
163 (34.2) 220 (46.2) 90 (18.9) 3 (0.6)
1 2 3 4
BI-RADS, breast imaging reporting and data system.
TABLE 3. Table Showing Correlation Between Results of Automated Volumetric Breast Density Calculation and BI-RADS Grading (By Agreement of Both Observers) BI-RADS Grade (By Agreement of Both Observers) 1 2 3 4
Volume of Fibroglandular Tissue
Breast Volume
Volumetric Breast Density
Number
P Value <0.001
P Value <0.001
P Value <0.001
156 202 83 3
2.664 ± 3.764 4.336 ± 3.261 5.858 ± 2.860 8.500 ± 6.477
47.933 ± 24.311 74.391 ± 26.004 50.055 ± 23.286 32.617 ± 14.167
3.853 ± 1.635 8.268 ± 4.410 12.432 ± 4.606 23.417 ± 9.994
TABLE 4. Results of Automated Volumetric Breast Density Calculation According to Age of Patients
Age <40 41–50 51–60 >60
Volume of Fibroglandular Tissue
Breast Volume
Volumetric Breast Density
Number
P Value <0.001
P Value <0.001
P Value <0.001
49 254 141 32
4.981 ± 3.208 4.641 ± 3.616 3.281 ± 1.692 2.227 ± 1.051
52.043 ± 26.701 71.420 ± 23.195 62.180 ± 25.121 58.364 ± 25.138
10.510 ± 5.336 8.799 ± 5.573 5.945 ± 3.624 4.389 ± 2.561
all BI-RADS categories (P value <0.001). Volumetric breast density increased with increased BI-RADS category. Mean fibroglandular volume and volumetric breast density was also calculated according to age range of the patients and results have been summarized in Table 4. Mean fibroglandular volume and breast density decreased with age of the patients. A negative correlation was found between age of the patients and VDG grade (ρ = −0.374). Using spearman rank coefficient, a positive correlation was found between VDG grading by software and BI-RADS categories (ρ = 0.728 for first observer and ρ = 0.725 for second observer). DISCUSSION Increased breast density is a risk factor for developing breast cancer. Besides increasing risk for breast cancer, breast density also has influence upon certain prognostic factors (like tumor size, lymph node status, hormone receptor status, etc.) (8). Thus, breast density could have a potential role in risk stratification as well as prognostication of breast cancer. Without 4
exact quantification, breast density cannot prove useful in clinical trials and risk assessment (22). BI-RADS is the most commonly used method for classifying mammographic density. However, classification based on visual analysis makes this method subjective and poorly reproducible (9–11). For more precise measurement of breast density, software for quantitative measurement of breast density have been developed, which are area based or volume based. One of the commonly used method for the breast density quantification on mammography is area-based, computer-assisted threshold method (13). However, this method does not take tissue thickness into account and is based upon threshold manually set by the observer. On the other hand, automated volumetric assessment of breast density has absolute reproducibility. Also, breast density estimates from FFDM examinations are more accurate than film-screen system as the calibration data (used for volumetric estimates) from FFDM system are more reliable than those obtained from film-screen system. Also in FFDM, the pixel value is related linearly to exposure. Thus, breast density estimates from FFDM are expected to be more reproducible and accurate than film-screen mammography (23).
Academic Radiology, Vol ■, No , ■■ 2015
In this study, we used fully automated volumetric software to calculate breast density from routine screening mammography. A total of 476 patients were analyzed using this software and VDG was provided from 1 to 4 according to volumetric breast density. In our study, a positive corelation was found between BI-RADS grading and VDG grading. Volumetric breast density increased with increasing BI-RADS category and there was a significant difference between volumetric breast density in each category. The results are in accordance with various previous studies using the same software (23–26). Tagliafico et al. (27) also reported good correlation between BI-RADS categories and breast density estimates with fully automated software (area-based density estimates) (r = 0.62, P < 0.01). In addition, breast density estimates from FFDM using fully automated software have shown good correlation with breast density estimates from magnetic resonance imaging in previous studies (28–30). Mammographic density measurement using area-based method (Cumulus) has also shown good correlation with density estimates from fully automated volumetric software (31). In our study, mean volumetric breast density (7.676 ± 4.891%) was less than the mean volumetric breast density (15.4 ± 7.8%) reported in a Korean study (21). Mean fibroglandular volume in our cohort (3.999 ± 2.529 cm3) was also less than the Korean cohort (51 ± 24.7 cm3). There was a significant difference in the number of patients in VDG grading and in BI-RADS grading. Using fully automated software, ~41% of patients were categorized as VDG grade 3 and grade 4. On the other hand, BI-RADS grade 3 and grade 4 was given to 19.5% of patients according to observer 1 and to 22.1% of patients according to observer 2. Gweon et al. also reported that more mammographic examinations were classified as grade 4 (41.6%) using the fully automated software than the three observers (20.2%) (23). Ko et al. reported that 45.7% of total mammograms were discordant in breast density evaluation by radiologists and by automated software (26). Out of these discordant cases in 423 (81.5%) mammograms, density was overestimated by automated volumetric software. Similarly, Seo et al. also reported that out of 59 (30.6%) discordant cases, 54 were overscored by VDG than visual assessment (24). Exact cause for this overestimation is not clear. This discordance may indicate greater sensitivity of the automated software to detect density in areas of the mammogram that appear less dense to human eye. Alternatively, discordance could also indicate a lower level of resolution by the automated software than by visual assessment (26). It has also been assumed that because the BI-RADS’ visual assessment is done on postprocessed images and the automated estimation is done on raw images, it can be responsible for discordance (24). Also, in one of the studies, total breast volume was significantly lower in the discordant group than in the concordant group (26). As will be discussed later, lower breast volume can also lead to increased VBD measurements. Also, in our study, breast volume was significantly decreased in BIRADS category 4. Gweon et al. suggested that the
BREAST DENSITY ESTIMATION
overestimation of breast density can be decreased by a correction in the mapping between average volumetric breast density and VDGs 1–4 (23). In our study, there was almost perfect agreement between the two observers for assigning BI-RADS cathegories. Gweon et al. (23) reported moderate interobserver agreement (κ: 0.48) among three radiologists’ estimates of BI-RADS density categories. Previous studies have also shown unsatisfactory interobserver aggreement (9–11). However, Seo et al. also reported good interobserver agreement among three radiologists for BI-RADS density category estimation with intraclass correlation coefficient of 0.860 (24). In yet another study comparing interobserver agreement among six radiologists in pairs, almost perfect agreement was seen between two pairs (κ: >0.8) (32). Also, both of our radiologists had expertise in women imaging. This could be one of the reasons for almost perfect interobserver agreement. As shown in Table 3, in BI-RADS category 4, mean fibroglandular volume and breast density were significantly increased and breast volume was significantly decreased. As the breast volume can have an effect on breast density, the automated software may show higher density estimates for smaller breasts. On the other hand, the assignment of BIRADS density categories by radiologists is not affected by breast size. Further study will be needed to evaluate the effect of breast size on breast density and breast cancer risk. We also analyzed mean volumetric breast density according to age range of the patients. There was a statistically significant difference between mean volumetric breast density and age of the patients. A negative correlation was found between mean volumetric breast density and age of the patients and volumetric breast density decreased with increasing age range of patients. Amanda et al. also reported inverse relationship between age and percent breast density measured from different quantitative methods (33). Our study had certain limitations. First, this was a small study done in a single institution and examinations were interpreted by only two radiologists. For validation of results, larger studies need to be performed on larger populations in different clinical settings involving a number of radiologists. Secondly, there is no reference standard for measurement of breast study. To overcome this limitation, we assumed the BI-RADS categories assigned by agreement of both the radiologists. Thirdly, the factors affecting mean volumetric density (other than age) were not investigated. A high-percentage breast density was strongly related to younger age, lower body mass index, nulliparity, late age at first delivery, and pre- and perimenopausal status (2,34). A few other studies have reported that high body mass index was related to a larger dense volume on volumetric methods (35,36). Finally, the present study included only an asymptomatic population with no abnormality on mammography. Thus, effect of any abnormality like mass, scar, etc. on volumetric breast density could not be investigated. Moreover, our study was done on a single mammography machine; thus, our results cannot be generalized to all types of machines. 5
SINGH ET AL
In conclusion, a positive correlation was found between fully automated volumetric estimation of breast density with BI-RADS density categories assigned by radiologists, with fair agreement between BI-RADS and VDG grading. Thus, fully automated volumetric method could be used for quantification of breast density. However, further studies are required to evaluate the causes of disagreement between visual assessment and automated density measurement. REFERENCES 1. Mainiero MB, Lourenco A, Mahoney MC, et al. ACR Appropriateness Criteria breast cancer screening. J Am Coll Radiol 2013; 10:11–14. 2. Boyd NF, Guo H, Martin LJ, et al. Mammographic density and the risk and detection of breast cancer. N Engl J Med 2007; 356:227–236. 3. Van Gils C, Otten JD, Verbeek AL, et al. Mammographic breast density and risk of breast cancer: masking bias or causality? Eur J Epidemiol 1998; 14:315–320. 4. Mandelson MT, Oestreicher N, Porter PL, et al. Breast density as a predictor of mammographic detection: comparison of interval- and screendetected cancers. J Natl Cancer Inst 2000; 92:1081–1087. Review. 5. Boyd NF, Martin LJ, Yaffe MJ, et al. Mammographic density and breast cancer risk: current understanding and future prospects. Breast Cancer Res 2011; 13:223. 6. McCormack VA, dos Santos Silva I. Breast density and parenchymal patterns as markers of breast cancer risk: a meta-analysis. Cancer Epidemiol Biomarkers Prev 2006; 15:1159–1169. 7. Vachon CM, van Gils CH, Sellers TA, et al. Mammographic density, breast cancer risk and risk prediction. Breast Cancer Res 2007; 9:217. 8. Bertrand KA, Tamimi RM, Scott CG, et al. Mammographic density and risk of breast cancer by age and tumour characteristics. Breast Cancer Res 2013; 15:R104. doi:10.1186/bcr3570. 9. Kerlikowske K, Grady D, Barclay J, et al. Variability and accuracy in mammographic interpretation using the American College of Radiology Breast Imaging Reporting and Data System. J Natl Cancer Inst 1998; 90:1801– 1809. 10. Berg WA, Campassi C, Langenberg P, et al. Breast Imaging Reporting and Data System: inter- and intraobserver variability in feature analysis and final assessment. AJR Am J Roentgenol 2000; 174:1769–1777. 11. Martin KE, Helvie MA, Zhou C, et al. Mammographic density measured with quantitative computer-aided method: comparison with radiologists’ estimates and BI-RADS categories. Radiology 2006; 240:656–665. 12. Spayne MC, Gard CC, Skelly J, et al. Reproducibility of BI-RADS breast density measures among community radiologists: a prospective cohort study. Breast J 2012; 18:326–333. 13. Byng JW, Boyd NF, Fishell E, et al. The quantitative analysis of mammographic densities. Phys Med Biol 1994; 39:1629–1638. 14. Sivaramakrishna R, Obuchowski NA, Chilcote WA, et al. Automatic segmentation of mammographic density. Acad Radiol 2001; 8:250–256. 15. Prevrhal S, Shepherd J, Smith-Bindman R, et al. Accuracy of mammographic breast density analysis: results of formal operator training. Cancer Epidemiol Biomarkers Prev 2002; 11:1389–1393. 16. Shepherd JA, Kerlikowske K, Ma L, et al. Volume of mammographic density and risk of breast cancer. Cancer Epidemiol Biomarkers Prev 2011; 20:1473–1482. 17. Highnam R, Pan X, Warren R, et al. Breast composition measurements using retrospective standard mammogram form (SMF). Phys Med Biol 2006; 51:2695–2713.
6
Academic Radiology, Vol ■, No , ■■ 2015
18. Jeffreys M, Warren R, Highnam R, et al. Initial experiences of using an automated volumetric measure of breast density: the standard mammogram form. Br J Radiol 2006; 79:378–382. 19. Highnam RP, Brady JM, Shepstone BJ. Mammographic image analysis. Eur J Radiol 1997; 24:20–32. 20. D’Orsi CJ, Sickles EA, Mendelson EB, et al. ACR BI-RADS—breast imaging atlas, breast imaging reporting and data system. Reston, VA: American College of Radiology, 2013. 21. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977; 33:159–174. 22. Harvey JA, Bovbjerg VE. Quantitative assessment of mammographic breast density: relationship with breast cancer risk. Radiology 2004; 230:29– 41. 23. Gweon HM, Youk JH, Kim JA, et al. Radiologist assessment of breast density by BI-RADS categories versus fully automated volumetric assessment. AJR Am J Roentgenol 2013; 201:692–697. 24. Seo JM, Ko ES, Han BK, et al. Automated volumetric breast density estimation: a comparison with visual assessment. Clin Radiol 2013; 68:690– 695. 25. McEntee MF, Damases CN. Mammographic density measurement: a comparison of automated volumetric density measurement to BIRADS. Proc SPIE Int Soc Opt Eng 2014; doi:10.1117/12.2042966. 26. Ko SY, Kim EK, Kim MJ, et al. Mammographic density estimation with automated volumetric breast density measurement. Korean J Radiol 2014; 15:313–321. 27. Tagliafico A, Tagliafico G, Tosto S, et al. Mammographic density estimation: comparison among BI-RADS categories, a semi-automated software and a fully automated one. Breast 2009; 18:35–40. 28. Gubern-Mérida A, Kallenberg M, Platel B, et al. Volumetric breast density estimation from full field digital mammograms: a validation study. PLoS ONE 2014; 9:e85952. doi:10.1371/journal.pone.0085952; eCollection 2014. 29. Wang J, Aziz A, Newitt D, et al. Comparison of Hologic’s Quantra volumetric assessment to MRI breast density. In: Proceedings of the 11th International Conference on Breast Imaging, IWDM‘12. Berlin, Heidelberg: Springer-Verlag., 2012; 619–626. 30. van Engeland S, Snoeren PR, Huisman H, et al. Volumetric breast density estimation from full-field digital mammograms. IEEE Trans Med Imaging 2006; 25:273–282. 31. Jeffreys M, Harvey J Highnam R. Comparing a new volumetric breast density method (Volpara™) to Cumulus. In: Martı’ J, ed. IWDM’10: Proceedings of the 10th International Workshop on Digital Mammography. Berlin, Heidelberg: Springer-Verlag, 2010; 408–413. 32. Bernardi D1, Pellegrini M, Di Michele S, et al. Interobserver agreement in breast radiological density attribution according to BI-RADS quantitative classification. Radiol Med 2012; 117:519–528. 33. Eng A, Gallant Z, Shepherd J, et al. Digital mammographic breast density and breast cancer risk: a case control study of six alternative density assessment methods. Breast Cancer Res 2014; 16:439. 34. Stone J, Warren RM, Pinney E, et al. Determinants of percentage and area measures of mammographic density. Am J Epidemiol 2009; 170:1571–1578. 35. Aitken Z, McCormack VA, Highnam RP, et al. Screen-film mammographic density and breast cancer risk: a comparison of the volumetric standard mammogram form and the interactive threshold measurement methods. Cancer Epidemiol Biomarkers Prev 2010; 19:418– 428. 36. Jeffreys M, Warren R, Highnam R, et al. Breast cancer risk factors and a novel measure of volumetric breast density: cross-sectional study. Br J Cancer 2008; 98:210–216.
Original Investigation
Breast Density Estimation with Fully Automated Volumetric Method: Comparison to Radiologists’ Assessment by BI-RADS Categories Tulika Singh, MD, Madhurima Sharma, MD, Veenu Singla, MD, Niranjan Khandelwal, MD, DNB Rationale and Objectives: The objective of our study was to calculate mammographic breast density with a fully automated volumetric breast density measurement method and to compare it to breast imaging reporting and data system (BI-RADS) breast density categories assigned by two radiologists. Materials and Methods: A total of 476 full-field digital mammography examinations with standard mediolateral oblique and craniocaudal views were evaluated by two blinded radiologists and BI-RADS density categories were assigned. Using a fully automated software, mean fibroglandular tissue volume, mean breast volume, and mean volumetric breast density were calculated. Based on percentage volumetric breast density, a volumetric density grade was assigned from 1 to 4. Results: The weighted overall kappa was 0.895 (almost perfect agreement) for the two radiologists’ BI-RADS density estimates. A statistically significant difference was seen in mean volumetric breast density among the BI-RADS density categories. With increased BIRADS density category, increase in mean volumetric breast density was also seen (P < 0.001). A significant positive correlation was found between BI-RADS categories and volumetric density grading by fully automated software (ρ = 0.728, P < 0.001 for first radiologist and ρ = 0.725, P < 0.001 for second radiologist). Pairwise estimates of the weighted kappa between Volpara density grade and BI-RADS density category by two observers showed fair agreement (κ = 0.398 and 0.388, respectively). Conclusions: In our study, a good correlation was seen between density grading using fully automated volumetric method and density grading using BI-RADS density categories assigned by the two radiologists. Thus, the fully automated volumetric method may be used to quantify breast density on routine mammography. Key Words: Automated; BI-RADS; Breast density; Mammography; Volumetric method. © 2015 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved.
INTRODUCTION
M
ammography is the recommended screening method for breast cancer in the general population (1). However, the sensitivity of screening mammography is reduced in women with dense breasts because of obscuration of underlying lesion by dense and heterogenous breast tissue. Because of this, chances of missing breast cancer are increased (2–4). At the same time, increased breast density has been considered as an independent risk factor for the development of breast cancer. Various studies have shown that women with high breast density compared to women with low breast density are four to six times more likely to get breast cancer (5–7). Bertrand et al. reported that women with breast Acad Radiol 2015; ■:■■–■■ From the Department of Radiodiagnosis and Imaging, Post Graduate Institute of Medical Education and Research (PGIMER), Chandigarh, India. Received July 3, 2015; revised September 17, 2015; accepted September 20, 2015. Address for correspondence to: T.S. e-mail: [email protected] © 2015 The Association of University Radiologists. Published by Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.acra.2015.09.012
density more than 50% have approximately twofold risk of developing breast cancer than women with breast density of 11–23%. The authors also reported a positive correlation between breast density and larger tumor size, positive lymph nodes, and negative estrogen receptor status (8). The American College of Radiology (ACR) recommends annual screening mammography in women more than 40 years of age. However, in a subset of population having increased risk for breast cancer, ACR not only recommends screening at earlier age but also emphasizes the need of supplemental screening methods like ultrasound and magnetic resonance imaging (1). ACR has also recommended supplemental screening in women with genetic predisposition to the disease and in women with dense breasts (1). Thus, breast density determines the need for supplemental screening and can prove to be an important factor in risk stratification. Various qualitative and quantitative methods have been devised for measurement of breast density on mammography. Breast imaging reporting and data system (BI-RADS) classification system is the commonest method in practice for measurement of breast density (9). The system is qualitative 1
SINGH ET AL
and based on visual analysis. Increased score implies increased breast density. However, this method is not reliably reproducible. Various studies done in the past have shown variable interobserver agreement among the radiologists (κ = 0.43–0.76) (10–12). In one large cohort study, wide variability was found among community radiologists in measurement of breast density and only 18% had moderate to substantial agreement (12). To overcome this problem, various quantitative methods for breast density measurement have been developed: visual computer assisted (semiautomated) or fully automated. These methods can be area based or volume based. Area-based method estimates breast density from area segmentation of two-dimensional image of compressed breast (13,14). The semiautomated method is based upon user-defined threshold thereby increasing subjectivity of the method and the results are highly user dependent, with substantial inter- and intraobserver variability (11,15). In addition, both qualitative and area-based methods have certain other limitations, including two-dimensional measurements, variability with breast compression, and no consideration of breast thickness (16). Because of these limitations, various volumetric methods have been developed for measuring breast density (17,18). Fully automated volumetric methods for density measurement are based upon physical model that assumes that the breast is composed of fibroglandular parenchyma and fat. Tissue composition in a given pixel can be calculated from x-ray attenuation properties of these tissues, and fibroglandular density can thus be estimated (18,19). Volumetric method takes breast thickness into account and is more reproducible. Limited number of studies has been carried out in the past to compare various methods of breast density estimation. The purpose of our study was to compare qualitative method of breast density grading using BI-RADS to breast density estimates from fully automated volumetric software. Moreover, most of the previous studies have been done on a Korean population. The objective of our study was to validate the results of fully automated volumetric method on an Indian cohort.
MATERIALS AND METHODS Patients and Mammography
This was a retrospective study approved by our institutional ethical committee. Informed written consent was obtained from all the patients. From March 2013 to April 2014, a total of 715 full-field digital mammography (FFDM) examinations with standard views (mediolateral oblique and craniocaudal) were performed. All examinations performed on asymptomatic females more than 35 years of age were included in this study. Patients with age less than 35 years, with breast symptoms, and with previous history of any breast surgery were excluded from the study. A total of 239 patients were excluded from the study. All digital mammographic examinations were performed on an FFDM unit: MicroDose SI, Philips (Amsterdam, Netherlands). The system employs crystalline silicon 2
Academic Radiology, Vol ■, No , ■■ 2015
photon counting chamber type of detector and has spatial resolution of 50 μm and image data matrix size of 4800 × 5200 pixels. BI-RADS Density Classification The mammography images were analyzed independently by two blinded radiologists who specialize in breast imaging and had 5–10 years of experience in interpreting mammography. All the images were reviewed in Digital imaging and communication (DICOM) in medicine format at the review workstation. The images were read as a batch at the end of the study. The radiologists assessed breast density in each mammogram according to the BI-RADS breast density categories. Mediolateral oblique and craniocaudal views were read simultaneously to assess the breast density. The following BIRADS categories for breast density were used for mammographic interpretations: category 1, almost fatty; category 2, scattered fibroglandular densities; category 3, heterogeneously dense; and category 4, extremely dense (20). Mammographic Density Analysis by Fully Automated Volumetric Software
For fully automated volumetric analysis, Volpara software (version 1.4.5, Ma¯takina Technology, LTD, Wellington, New Zealand) was used. The Volpara software processes the image data generated by the digital mammography system and calculates the breast density. The measurement starts with finding a reference point in the breast having known composition (usually near the chest wall containing all fatty tissue). Then x-ray attenuation is calculated in each pixel. From degree of attenuation in a pixel, composition of tissue located between the pixel and the x-ray source is estimated and density maps are created. By analyzing values in density maps, the software computes the fibroglandular tissue volume, the breast volume (both in cubic centimetres), and the volumetric breast density. The volumetric density is then computed from these data, which range from 0% to 40%. Mediolateral oblique and craniocaudal views are averaged and the density information is provided per breast. A Volpara density grade (VDG) is also given to each patient. Total fibroglandular tissue volume is divided by total breast volume to obtain a percentage per patient. The resultant percentage is graded as VDG as follows: 0–4.7% volumetric density, VDG 1; 4.8–7.9%, VDG 2; 8.0– 15.0%, VDG 3; and 15.1% and above, VDG 4. After completion of the study, Volpara automatically processes data and sends DICOM secondary capture images containing the breast density information (Fig. 1). Data and Statistical Analysis
Interobserver agreement in measurement of breast density using BI-RADS was calculated using weighted kappa. The kappa values were interpreted as suggested by Landis and Koch (21) as follows: a kappa value equal to or less than 0.20 indicates
Academic Radiology, Vol ■, No , ■■ 2015
BREAST DENSITY ESTIMATION
TABLE 1. Table Showing Age of the Patients and Various Parameters Obtained by Automated Breast Density Calculation Using Volpara Software (Version 1.4.5) Age (mean ± SD) Volume of fibroglandular tissue (mean ± SD) Breast volume (mean ± SD) Volumetric breast density (mean ± SD)
48.8 ± 7.07 3.999 ± 2.529 57.905 ± 25.325 7.676 ± 4.891
Figure 1. Image showing example of breast density estimation using fully automated volumetric method (Volpara version 1.4.5).
Figure 3. 1–4.
Figure 2. Age distribution of patients included in the study.
slight agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, almost perfect agreement. Statistical comparisons were performed with independent and paired Student t tests for continuous variables and the chi-square or Fisher exact test for categorical variables. The correlation between the BIRADS density category and the volumetric breast density provided by the fully automated software was estimated using the Spearman rank correlation coefficient (ρ). Statistical analysis was performed using SPSS Statistics 17. Differences were considered to be statistically significant at P < 0.05. RESULTS A total of 476 patients were enrolled in the study. The age range of patients was from 36 to 76 years with a mean of 48.8 years and a standard deviation of 7.07 years (Fig. 2). Most of the patients were in the age range of 41–50 years (53.4%) and 51–60 years (29.6%). Using fully automated software (Volpara, version 1.4.5), mean volume of fibroglandular tissue, breast volume, and volumetric breast density were derived and have been summarized
Bar chart showing frequency of patients in VDG grade
in Table 1. According to volumetric density, VDG grade from 1 to 4 was assigned (Fig. 3). Interobserver agreement between two radiologists for the assignment of BI-RADS categories was almost perfect (κ = 0.895). Pairwise estimates of the weighted kappa between VDG grade and BI-RADS density category by two observers showed fair agreement (κ = 0.398 and 0.388, respectively). On visual assessment, less than one fourth of the study population was categorized as BI-RADS 3 or BI-RADS 4 (Table 2). According to the results from the first radiologist’s grading of BI-RADS density category, BI-RADS category 3 was found in 20.6% of the cases and BI-RADS category 4 was found in 1.5% of the cases. Similarly, according to the second radiologist, 18.9% of the patients were assigned BI-RADS category 3 and 0.6% of the patients were assigned BIRADS category 4 (Table 2). However, automated volumetric assessment from the Volpara software showed ~41% of the patients having VDG grade 3 (34.5%) and grade 4 (6.9%) (Fig. 3). Out of 476 examinations, 444 examinations (93.3%) showed agreement between BI-RADS grading of both the radiologists. In the remaining 32 patients with disagreement, differences were within one category only. Mean volumetric breast density of all patients in each BI-RADS category was also calculated (by agreement of both radiologists) (Table 3). There was a significant difference between volumetric breast density of 3
Academic Radiology, Vol ■, No , ■■ 2015
SINGH ET AL
TABLE 2. Table Showing Number of Patients in Each Grade According to Volpara Density Grade (VDG) and BI-RADS Grading by Two Observers
VDG Grade
Number of Patients (Percentage)
BI-RADS Grade
Observer 1
Observer 2
148 (31.1) 131 (27.5) 164 (34.5) 33 (6.9)
1 2 3 4
159 (33.4) 212 (44.5) 98 (20.6) 7 (1.5)
163 (34.2) 220 (46.2) 90 (18.9) 3 (0.6)
1 2 3 4
BI-RADS, breast imaging reporting and data system.
TABLE 3. Table Showing Correlation Between Results of Automated Volumetric Breast Density Calculation and BI-RADS Grading (By Agreement of Both Observers) BI-RADS Grade (By Agreement of Both Observers) 1 2 3 4
Volume of Fibroglandular Tissue
Breast Volume
Volumetric Breast Density
Number
P Value <0.001
P Value <0.001
P Value <0.001
156 202 83 3
2.664 ± 3.764 4.336 ± 3.261 5.858 ± 2.860 8.500 ± 6.477
47.933 ± 24.311 74.391 ± 26.004 50.055 ± 23.286 32.617 ± 14.167
3.853 ± 1.635 8.268 ± 4.410 12.432 ± 4.606 23.417 ± 9.994
TABLE 4. Results of Automated Volumetric Breast Density Calculation According to Age of Patients
Age <40 41–50 51–60 >60
Volume of Fibroglandular Tissue
Breast Volume
Volumetric Breast Density
Number
P Value <0.001
P Value <0.001
P Value <0.001
49 254 141 32
4.981 ± 3.208 4.641 ± 3.616 3.281 ± 1.692 2.227 ± 1.051
52.043 ± 26.701 71.420 ± 23.195 62.180 ± 25.121 58.364 ± 25.138
10.510 ± 5.336 8.799 ± 5.573 5.945 ± 3.624 4.389 ± 2.561
all BI-RADS categories (P value <0.001). Volumetric breast density increased with increased BI-RADS category. Mean fibroglandular volume and volumetric breast density was also calculated according to age range of the patients and results have been summarized in Table 4. Mean fibroglandular volume and breast density decreased with age of the patients. A negative correlation was found between age of the patients and VDG grade (ρ = −0.374). Using spearman rank coefficient, a positive correlation was found between VDG grading by software and BI-RADS categories (ρ = 0.728 for first observer and ρ = 0.725 for second observer). DISCUSSION Increased breast density is a risk factor for developing breast cancer. Besides increasing risk for breast cancer, breast density also has influence upon certain prognostic factors (like tumor size, lymph node status, hormone receptor status, etc.) (8). Thus, breast density could have a potential role in risk stratification as well as prognostication of breast cancer. Without 4
exact quantification, breast density cannot prove useful in clinical trials and risk assessment (22). BI-RADS is the most commonly used method for classifying mammographic density. However, classification based on visual analysis makes this method subjective and poorly reproducible (9–11). For more precise measurement of breast density, software for quantitative measurement of breast density have been developed, which are area based or volume based. One of the commonly used method for the breast density quantification on mammography is area-based, computer-assisted threshold method (13). However, this method does not take tissue thickness into account and is based upon threshold manually set by the observer. On the other hand, automated volumetric assessment of breast density has absolute reproducibility. Also, breast density estimates from FFDM examinations are more accurate than film-screen system as the calibration data (used for volumetric estimates) from FFDM system are more reliable than those obtained from film-screen system. Also in FFDM, the pixel value is related linearly to exposure. Thus, breast density estimates from FFDM are expected to be more reproducible and accurate than film-screen mammography (23).
Academic Radiology, Vol ■, No , ■■ 2015
In this study, we used fully automated volumetric software to calculate breast density from routine screening mammography. A total of 476 patients were analyzed using this software and VDG was provided from 1 to 4 according to volumetric breast density. In our study, a positive corelation was found between BI-RADS grading and VDG grading. Volumetric breast density increased with increasing BI-RADS category and there was a significant difference between volumetric breast density in each category. The results are in accordance with various previous studies using the same software (23–26). Tagliafico et al. (27) also reported good correlation between BI-RADS categories and breast density estimates with fully automated software (area-based density estimates) (r = 0.62, P < 0.01). In addition, breast density estimates from FFDM using fully automated software have shown good correlation with breast density estimates from magnetic resonance imaging in previous studies (28–30). Mammographic density measurement using area-based method (Cumulus) has also shown good correlation with density estimates from fully automated volumetric software (31). In our study, mean volumetric breast density (7.676 ± 4.891%) was less than the mean volumetric breast density (15.4 ± 7.8%) reported in a Korean study (21). Mean fibroglandular volume in our cohort (3.999 ± 2.529 cm3) was also less than the Korean cohort (51 ± 24.7 cm3). There was a significant difference in the number of patients in VDG grading and in BI-RADS grading. Using fully automated software, ~41% of patients were categorized as VDG grade 3 and grade 4. On the other hand, BI-RADS grade 3 and grade 4 was given to 19.5% of patients according to observer 1 and to 22.1% of patients according to observer 2. Gweon et al. also reported that more mammographic examinations were classified as grade 4 (41.6%) using the fully automated software than the three observers (20.2%) (23). Ko et al. reported that 45.7% of total mammograms were discordant in breast density evaluation by radiologists and by automated software (26). Out of these discordant cases in 423 (81.5%) mammograms, density was overestimated by automated volumetric software. Similarly, Seo et al. also reported that out of 59 (30.6%) discordant cases, 54 were overscored by VDG than visual assessment (24). Exact cause for this overestimation is not clear. This discordance may indicate greater sensitivity of the automated software to detect density in areas of the mammogram that appear less dense to human eye. Alternatively, discordance could also indicate a lower level of resolution by the automated software than by visual assessment (26). It has also been assumed that because the BI-RADS’ visual assessment is done on postprocessed images and the automated estimation is done on raw images, it can be responsible for discordance (24). Also, in one of the studies, total breast volume was significantly lower in the discordant group than in the concordant group (26). As will be discussed later, lower breast volume can also lead to increased VBD measurements. Also, in our study, breast volume was significantly decreased in BIRADS category 4. Gweon et al. suggested that the
BREAST DENSITY ESTIMATION
overestimation of breast density can be decreased by a correction in the mapping between average volumetric breast density and VDGs 1–4 (23). In our study, there was almost perfect agreement between the two observers for assigning BI-RADS cathegories. Gweon et al. (23) reported moderate interobserver agreement (κ: 0.48) among three radiologists’ estimates of BI-RADS density categories. Previous studies have also shown unsatisfactory interobserver aggreement (9–11). However, Seo et al. also reported good interobserver agreement among three radiologists for BI-RADS density category estimation with intraclass correlation coefficient of 0.860 (24). In yet another study comparing interobserver agreement among six radiologists in pairs, almost perfect agreement was seen between two pairs (κ: >0.8) (32). Also, both of our radiologists had expertise in women imaging. This could be one of the reasons for almost perfect interobserver agreement. As shown in Table 3, in BI-RADS category 4, mean fibroglandular volume and breast density were significantly increased and breast volume was significantly decreased. As the breast volume can have an effect on breast density, the automated software may show higher density estimates for smaller breasts. On the other hand, the assignment of BIRADS density categories by radiologists is not affected by breast size. Further study will be needed to evaluate the effect of breast size on breast density and breast cancer risk. We also analyzed mean volumetric breast density according to age range of the patients. There was a statistically significant difference between mean volumetric breast density and age of the patients. A negative correlation was found between mean volumetric breast density and age of the patients and volumetric breast density decreased with increasing age range of patients. Amanda et al. also reported inverse relationship between age and percent breast density measured from different quantitative methods (33). Our study had certain limitations. First, this was a small study done in a single institution and examinations were interpreted by only two radiologists. For validation of results, larger studies need to be performed on larger populations in different clinical settings involving a number of radiologists. Secondly, there is no reference standard for measurement of breast study. To overcome this limitation, we assumed the BI-RADS categories assigned by agreement of both the radiologists. Thirdly, the factors affecting mean volumetric density (other than age) were not investigated. A high-percentage breast density was strongly related to younger age, lower body mass index, nulliparity, late age at first delivery, and pre- and perimenopausal status (2,34). A few other studies have reported that high body mass index was related to a larger dense volume on volumetric methods (35,36). Finally, the present study included only an asymptomatic population with no abnormality on mammography. Thus, effect of any abnormality like mass, scar, etc. on volumetric breast density could not be investigated. Moreover, our study was done on a single mammography machine; thus, our results cannot be generalized to all types of machines. 5
SINGH ET AL
In conclusion, a positive correlation was found between fully automated volumetric estimation of breast density with BI-RADS density categories assigned by radiologists, with fair agreement between BI-RADS and VDG grading. Thus, fully automated volumetric method could be used for quantification of breast density. However, further studies are required to evaluate the causes of disagreement between visual assessment and automated density measurement. REFERENCES 1. Mainiero MB, Lourenco A, Mahoney MC, et al. ACR Appropriateness Criteria breast cancer screening. J Am Coll Radiol 2013; 10:11–14. 2. Boyd NF, Guo H, Martin LJ, et al. Mammographic density and the risk and detection of breast cancer. N Engl J Med 2007; 356:227–236. 3. Van Gils C, Otten JD, Verbeek AL, et al. Mammographic breast density and risk of breast cancer: masking bias or causality? Eur J Epidemiol 1998; 14:315–320. 4. Mandelson MT, Oestreicher N, Porter PL, et al. Breast density as a predictor of mammographic detection: comparison of interval- and screendetected cancers. J Natl Cancer Inst 2000; 92:1081–1087. Review. 5. Boyd NF, Martin LJ, Yaffe MJ, et al. Mammographic density and breast cancer risk: current understanding and future prospects. Breast Cancer Res 2011; 13:223. 6. McCormack VA, dos Santos Silva I. Breast density and parenchymal patterns as markers of breast cancer risk: a meta-analysis. Cancer Epidemiol Biomarkers Prev 2006; 15:1159–1169. 7. Vachon CM, van Gils CH, Sellers TA, et al. Mammographic density, breast cancer risk and risk prediction. Breast Cancer Res 2007; 9:217. 8. Bertrand KA, Tamimi RM, Scott CG, et al. Mammographic density and risk of breast cancer by age and tumour characteristics. Breast Cancer Res 2013; 15:R104. doi:10.1186/bcr3570. 9. Kerlikowske K, Grady D, Barclay J, et al. Variability and accuracy in mammographic interpretation using the American College of Radiology Breast Imaging Reporting and Data System. J Natl Cancer Inst 1998; 90:1801– 1809. 10. Berg WA, Campassi C, Langenberg P, et al. Breast Imaging Reporting and Data System: inter- and intraobserver variability in feature analysis and final assessment. AJR Am J Roentgenol 2000; 174:1769–1777. 11. Martin KE, Helvie MA, Zhou C, et al. Mammographic density measured with quantitative computer-aided method: comparison with radiologists’ estimates and BI-RADS categories. Radiology 2006; 240:656–665. 12. Spayne MC, Gard CC, Skelly J, et al. Reproducibility of BI-RADS breast density measures among community radiologists: a prospective cohort study. Breast J 2012; 18:326–333. 13. Byng JW, Boyd NF, Fishell E, et al. The quantitative analysis of mammographic densities. Phys Med Biol 1994; 39:1629–1638. 14. Sivaramakrishna R, Obuchowski NA, Chilcote WA, et al. Automatic segmentation of mammographic density. Acad Radiol 2001; 8:250–256. 15. Prevrhal S, Shepherd J, Smith-Bindman R, et al. Accuracy of mammographic breast density analysis: results of formal operator training. Cancer Epidemiol Biomarkers Prev 2002; 11:1389–1393. 16. Shepherd JA, Kerlikowske K, Ma L, et al. Volume of mammographic density and risk of breast cancer. Cancer Epidemiol Biomarkers Prev 2011; 20:1473–1482. 17. Highnam R, Pan X, Warren R, et al. Breast composition measurements using retrospective standard mammogram form (SMF). Phys Med Biol 2006; 51:2695–2713.
6
Academic Radiology, Vol ■, No , ■■ 2015
18. Jeffreys M, Warren R, Highnam R, et al. Initial experiences of using an automated volumetric measure of breast density: the standard mammogram form. Br J Radiol 2006; 79:378–382. 19. Highnam RP, Brady JM, Shepstone BJ. Mammographic image analysis. Eur J Radiol 1997; 24:20–32. 20. D’Orsi CJ, Sickles EA, Mendelson EB, et al. ACR BI-RADS—breast imaging atlas, breast imaging reporting and data system. Reston, VA: American College of Radiology, 2013. 21. Landis JR, Koch GG. The measurement of observer agreement for categorical data. Biometrics 1977; 33:159–174. 22. Harvey JA, Bovbjerg VE. Quantitative assessment of mammographic breast density: relationship with breast cancer risk. Radiology 2004; 230:29– 41. 23. Gweon HM, Youk JH, Kim JA, et al. Radiologist assessment of breast density by BI-RADS categories versus fully automated volumetric assessment. AJR Am J Roentgenol 2013; 201:692–697. 24. Seo JM, Ko ES, Han BK, et al. Automated volumetric breast density estimation: a comparison with visual assessment. Clin Radiol 2013; 68:690– 695. 25. McEntee MF, Damases CN. Mammographic density measurement: a comparison of automated volumetric density measurement to BIRADS. Proc SPIE Int Soc Opt Eng 2014; doi:10.1117/12.2042966. 26. Ko SY, Kim EK, Kim MJ, et al. Mammographic density estimation with automated volumetric breast density measurement. Korean J Radiol 2014; 15:313–321. 27. Tagliafico A, Tagliafico G, Tosto S, et al. Mammographic density estimation: comparison among BI-RADS categories, a semi-automated software and a fully automated one. Breast 2009; 18:35–40. 28. Gubern-Mérida A, Kallenberg M, Platel B, et al. Volumetric breast density estimation from full field digital mammograms: a validation study. PLoS ONE 2014; 9:e85952. doi:10.1371/journal.pone.0085952; eCollection 2014. 29. Wang J, Aziz A, Newitt D, et al. Comparison of Hologic’s Quantra volumetric assessment to MRI breast density. In: Proceedings of the 11th International Conference on Breast Imaging, IWDM‘12. Berlin, Heidelberg: Springer-Verlag., 2012; 619–626. 30. van Engeland S, Snoeren PR, Huisman H, et al. Volumetric breast density estimation from full-field digital mammograms. IEEE Trans Med Imaging 2006; 25:273–282. 31. Jeffreys M, Harvey J Highnam R. Comparing a new volumetric breast density method (Volpara™) to Cumulus. In: Martı’ J, ed. IWDM’10: Proceedings of the 10th International Workshop on Digital Mammography. Berlin, Heidelberg: Springer-Verlag, 2010; 408–413. 32. Bernardi D1, Pellegrini M, Di Michele S, et al. Interobserver agreement in breast radiological density attribution according to BI-RADS quantitative classification. Radiol Med 2012; 117:519–528. 33. Eng A, Gallant Z, Shepherd J, et al. Digital mammographic breast density and breast cancer risk: a case control study of six alternative density assessment methods. Breast Cancer Res 2014; 16:439. 34. Stone J, Warren RM, Pinney E, et al. Determinants of percentage and area measures of mammographic density. Am J Epidemiol 2009; 170:1571–1578. 35. Aitken Z, McCormack VA, Highnam RP, et al. Screen-film mammographic density and breast cancer risk: a comparison of the volumetric standard mammogram form and the interactive threshold measurement methods. Cancer Epidemiol Biomarkers Prev 2010; 19:418– 428. 36. Jeffreys M, Warren R, Highnam R, et al. Breast cancer risk factors and a novel measure of volumetric breast density: cross-sectional study. Br J Cancer 2008; 98:210–216.