BJR
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
British Journal of Radiology (2006) 79, 378-382
© 2006 British Institute of Radiology
doi: 10.1259/bjr/24769358
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Jeffreys, M
Right arrow Articles by Davey Smith, G
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Jeffreys, M
Right arrow Articles by Davey Smith, G

Full paper

Initial experiences of using an automated volumetric measure of breast density: the standard mammogram form

M Jeffreys, PhD 1 R Warren, FRCR 2 R Highnam, PhD 3 and G Davey Smith, DSc 4

1 Centre for Public Health Research, Massey University, Private Box 756, Wellington, New Zealand, 2 Addenbrooke's Hospital, Cambridge, 3 Mirada Solutions Ltd, Oxford, 4 Department of Social Medicine, University of Bristol, UK


   Abstract
Top
 Abstract
Introduction
 Materials and methods
 Results
 Discussion
 References
 
Limitations of area based measures of breast density have led several research groups to develop volumetric measures of breast density, for use in predicting risk and in epidemiological research. In this paper, we describe our initial experiences using an automated algorithm (standard mammogram form, SMF) to estimate the volume of the breast that is dense from digitized film mammograms. We performed analyses on 3816 mammograms of 626 women, who were part of the Glasgow Alumni Cohort and had mammograms taken within the Scottish Breast Screening Programme between 1989 and 2002. Absolute volume of dense breast tissue (SMF volume) and the percentage of the volume of the breast that is dense (SMF%) were calculated. The median (interquartile range) of SMF volume was 66 cm3 (48 to 98), and of SMF% was 23.4% (18.6 to 29.7). SMF%, but not SMF volume, was positively related to a six category classification (SCC) of visually assigned area-based breast density (increase in ln(SMF%) per category increase in SCC: 0.04% (95%CI: 0.03–0.05). The SMF algorithm produced lower SMF volume for craniocaudal (CC) compared with mediolateral oblique (MLO) views, but CC/MLO differences for SMF% were small. The mean right/left difference for ln(SMF volume) was –0.027 cm3 (95% confidence interval (CI) –0.044 to –0.009) and of ln(SMF%) was 0.005% (95% CI –0.008% to 0.019%). We present these initial data as a background for future analytical work using SMF.


   Introduction
Top
 Abstract
 Introduction
Materials and methods
 Results
 Discussion
 References
 
The magnitude of the relationship between breast density and breast cancer [1] has led to recognition that breast density may have potential use as a biomarker for breast cancer risk [2]. Breast density is conventionally estimated by using the area of the mammogram that appears to be "glandular" and dividing by the total breast area. This is performed either visually, using classifications such as the Wolfe system [3], or using a computer-based thresholding technique [1]. Each of these classifications results in a measure of breast density which is a strong determinant of breast cancer risk.

Several recognized breast cancer risk factors are positively related to area measures of breast density, including height [4, 5], parity [6] and age at first birth [7]. There are two notable exceptions to this, namely age and body weight, both of which are positively related to breast cancer risk, but inversely related to breast density [5, 8]. An inverse association between body mass index (BMI) and percent breast density is inevitable, since inherent in the definition of these measures is the assumption that fatty tissue is non-dense, and women with a high BMI have higher amounts of fat in the breasts. In addition to this discrepancy, there are other concerns associated with the visual methods of assessing breast density. These include: (i) the subjectivity of visual measures; (ii) variations in visual density with breast compression and X-ray exposure; (iii) consideration of area measures of the breast, despite its three-dimensional structure.

Due to these concerns regarding area-based measures, recent research has been directed towards volume-based measurements, which try to model the volume of glandular tissue. We describe here initial results from the Standard Mammogram Form (SMFTM) technology, version 2.2, a fully automated objective measurement tool to estimate the volume of glandular tissue in the breast from a mammogram [9, 10]. The SMF algorithm explicitly considers breast compression, exposure and tube voltage, and computes two volumetric measures of breast density, (i) the absolute volume (cm3) of the breast that is dense (SMF volume) and (ii) the percentage of the volume of the breast which is dense (SMF%). SMF is different from other volumetric research methods in that it incorporates a full physics model rather than using step-wedges in each image [11, 12].


   Materials and methods
Top
 Abstract
 Introduction
 Materials and methods
Results
 Discussion
 References
 
The data presented were obtained from women who are part of the Glasgow Alumni Cohort, which has been described in detail elsewhere [13]. The included women were students who were registered at the University of Glasgow during the period 1948–1968 and who attended an annual medical examination at the Student Health Service. Surviving cohort members were contacted by postal questionnaire in 2001. Those women who replied to the questionnaire and were living in Scotland were asked to give consent for access to screening mammograms taken under the Scottish Breast Screening Programme (SBSP) between 1989 and 2002. Women were informed that their mammograms would be digitized.

All screening mammograms for each woman were retrieved from the eight breast screening centres and were digitized on site with a Canon FS300 digitizer scanner at a resolution of 100 µm with 8-bit precision by a single radiographer. Data on exposure and tube voltage were collected. Both the postal questionnaire survey and the acquisition of digital mammograms received ethical approval from the Multi-centre Research Ethics Committee (Scotland).

For the visual assignment of density categories, scanned images were displayed at 300 µm resolution on a flat-panel display system. At this resolution, the images were about the same size as a mammogram film. All images were displayed to appear as if viewed on a light-box. No other adjustment or image post-processing was applied during the reading period. We have previously reported on the similarity in density measures obtained when these assessments are made from the digitized image compared with the original film [14]. Visual density measures were made by one radiologist experienced in density assessment (RW) using a six-point categorical scale of the percentage of the breast area that appeared dense. The categories were: 0%, 1–10%, 11–24%, 25–49%, 50–74% and ≥75% and are referred to in this paper as the six category classification (SCC). These scales were chosen to make our work comparable with that of other researchers [15]. All mammograms for each woman were presented consecutively to the radiologist.

The volume of the mammogram that appeared dense was estimated using SMF, v2.2. This is a computer algorithm which models the image formation process to decompose the breast into fatty and non-fatty tissue. It achieves this through estimating the thickness of dense tissue in each column of tissue between each pixel on the projected image and the X-ray source. The image is then standardized to remove the dependence on the parameters which were used to form the image. The input parameters required are side (left or right), view (craniocaudal (CC) or mediolateral oblique (MLO), current time product (mAs), tube voltage (kVp) and filter and target materials. The algorithm is fully automated, only requiring user intervention if there has been a data entry error, for example if a right sided mammogram was entered as left sided, since this causes the breast segmentation algorithm to fail. Detailed explanations of the physics behind the model have been published previously [9, 10].

Statistical analyses
Descriptive analyses were performed on all mammograms of all women. Mann-Whitney tests were used to test the differences in density measures obtained from left and right mammograms, and from different views (CC and MLO). Because of the log-normal distribution of the data, both SMF and SMF% were log transformed prior to analysis, and the natural log of these measures was used in the regression models. The new variables are referred to as ln(SMF volume) and ln(SMF%). Random effects linear regression models were used to estimate associations between age and SMF density measures, taking into account the clustered nature of the data (several mammograms per woman).

Further analysis was based on paired mammograms (i.e. left vs right, CC vs MLO) taken on the same day. For these, the first visit per woman was used, since it is at this visit that both views have been routinely performed since 1994, before two-view mammography became routine at all visits by 2003. Because of the continuous nature of the SMF data (both volume and percentage), Bland-Altman plots [16] of the natural log of the SMF measures were used instead of kappa statistics to assess the agreement between paired data. These plot the mean difference between pairs (expected to be zero) against the average value (ln(SMF volume) or ln(SMF%)) of that pair.


   Results
Top
 Abstract
 Introduction
 Materials and methods
 Results
Discussion
 References
 
There were 3566 women in the original Glasgow Alumni Cohort, of whom 2169 (61%) were sent a postal questionnaire in 2001. These were the women who could be traced through the National Health Service Central Register and were still alive. The response rate was 59% (n = 1285). Of the respondents, 935 women (73%) were still living in Scotland. 277 of these women (30%) had never had a screening mammogram, and two women refused access to their films.

The SMF algorithm was run on all 3968 mammograms belonging to 649 of the remaining 656 women (films of seven women were omitted inadvertently). The programme failed on one image and produced a result classified as "not excellent" for 29 (0.7%) further images. 23 (3.5%) further women (122 images) were excluded as they reported having had breast cancer in the 2001 questionnaire. Analyses are based on the remaining 626 women with 3816 mammograms.

The median age at first breast screening was 53.6 years (range 40.0–71.5 years). Eight women were over 65 years at the time of their first mammogram. The median (interquartile range (IQR)) of the absolute volume of SMF was 66 cm3 (48–98 cm3), and of SMF% was 23.4% (18.6–29.7%), see Figure 1Go.


Figure 1
View larger version (18K):
[in this window]
[in a new window]
  		Figure 1. Distribution of the volume of dense breast tissue in 3816 mammograms.

 
There was a non-linear relationship between the total volume of the breast and the volume of the breast which was dense, see Figure 2Go. In women with small overall breast volume, the absolute volume of SMF dense tissue was low, but the percentage of the volume of the breast which was dense was variable. In women with larger breast volume, the absolute volume of SMF dense tissue was variable, whereas the percentage of the volume of the breast which was dense tended to be smaller.


Figure 2
View larger version (33K):
[in this window]
[in a new window]
  		Figure 2. Relationship between total breast volume and standard mammogram form(SMF) volume/SMF% in 3816 mammograms.

 
Assessment of the association between the two SMF measures and the SCC system showed no relationship between absolute SMF volume and SCC, but a strong positive association between SMF% and SCC (Table 1Go). Each category increase in SCC was associated with a 0.04% (95%CI: 0.03–0.05) higher ln(SMF%). There were small but significant associations between each of the two SMF measures with age at mammography. The regression coefficient per year older, based on the ln(absolute SMF volume) was –0.008 cm3 (95%CI: –0.010 to –0.005). For ln(SMF%), the regression coefficient was –0.021% (–0.023 to –0.020).


View this table:
[in this window]
[in a new window]
  		Table 1. Association between area percent density and standard mammogram form(SMF) density in mammograms of 649 women

 
Comparisons of median SMF values stratified by mammogram view (CC vs MLO) indicated that the SMF algorithm consistently produced lower SMF volume for CC compared with MLO views (Table 2Go). The CC/MLO differences for SMF% were smaller. Comparing left and right sides, there was a small difference in the median absolute SMF volume, being slightly larger for left than right breasts, but no difference for SMF%.


View this table:
[in this window]
[in a new window]
  		Table 2. Association between view, side and standard mammogram form(SMF) density in mammograms of 649 women

 
Correlations between the paired mammograms taken on the same day were high. For mammograms taken on the same day, the left–right correlation was high, r = 0.92 (p<0.001) for ln(SMF volume) and r = 0.85 (p<0.001) for ln(SMF%). These correlations were equally high for CC and MLO views. Bland-Altman plots (Figure 3Go) showed that although left/right agreement was good for the majority of women, there were some women whose values lie well outside the mean reference range (±2 standard deviations of the mean difference between the left/right measures). The mean difference (right minus left) of the ln(SMF volume) was –0.027 cm3 (95% confidence interval (CI) –0.044 to –0.009) and of ln(SMF%) was 0.005% (95% confidence interval –0.008% to 0.019%).


Figure 3
View larger version (20K):
[in this window]
[in a new window]
  		Figure 3. Bland-Altman plots showing agreement between left/right pairs of mammograms, 626 women. Note: Bland-Altman plots of the difference against the mean of left and right measures, with horizontal lines showing the mean difference of 0 and limits of agreement (±2 standard deviations). Both standard mammogram form (SMF) and SMF% have been log transformed for these analyses.

 

   Discussion
Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
References
 
This study describes our initial experiences of using a new method to assess mammographic breast density, in which we estimate the proportion of the volume of the breast that is dense and the absolute volume of dense tissue. This new method is fully automated, and produced a useable density value for over 99% of images. We have shown that the percentage measure is closely associated with frequently used visual assessments of density. We had no gold standard against which to compare the absolute volume of density, but the good agreement between left and right SMF estimates provides an assurance regarding the internal consistency of the method.

Unlike the visual and semi-automated systems to assess breast density, which have been related to breast cancer in several studies [1], SMF has not yet been validated directly against the incidence of breast cancer. Combining a biological understanding of breast density with knowledge of the physics of mammogram acquisition, we believe that this is likely to be as strong a predictor of breast cancer as the more conventional methods of mammographic density assessment. Since the estimation of density using this method is entirely objective, we expect that the non-differential misclassification during visual estimation, which attenuates associations between density and breast cancer, will not occur. Associations between SMF and breast cancer may therefore be of greater magnitude than the four to six fold risks reported between extreme categories of visually estimated or computer-assisted methods [1].

A limitation of all currently proposed volumetric systems, including SMF, is that they fail to acknowledge that the non-fatty component consists of groups of fibrous, glandular and other tissues together. It has been suggested that the (possibly unconscious) acknowledgment of these structures by radiologists when assigning the visual percent density may explain the power of these assessments in determining breast cancer risk. Comparison of visual area-based density measures have shown that correlations between these and MR images are high [17], and of similar magnitude to the associations which we report. Although correlation coefficients are not recommended as measures of assessing agreement [16, 18], we present these results as a comparison with those reported in previous studies. For example, in a small sample of pre-menopausal women, Pearson correlation coefficients were between 0.86 and 0.96 [19], very similar to the results we present. Recent data suggest that the association between breast density and breast cancer risk is similar in pre- and post-menopausal women [20].

We found that the SMF algorithm appears to estimate higher SMF% values in CC compared with MLO mammograms, but lower SMF volume. This may be because the MLO, but not the CC, view includes the axillary tail, which is primarily composed of fat. Furthermore, the MLO image captures all the tissue at the very back of the breast, so has a higher SMF volume, but since that tissue is often mostly fatty, the SMF% for CC is higher.

If SMF can be proven in validated studies to be a useful marker of breast cancer risk, its potential is enormous. Unlike some methods of measuring volumetric breast density [11, 12], SMF does not require step wedges to be included during the mammogram acquisition. We have shown that it can be used with historical mammograms, which is a strong feature for use in epidemiological studies. Although SMF does currently require calibration data such as mAs and kVp to be known, work is well underway to remove this requirement (RH, personal communication, 2005). Development that would enable volume measurements of this kind without the need for calibration data would enable wider use for epidemiology in a multicentre setting.

Despite continuing reported associations between risk factors, breast density and breast cancer [2, 20], there is a need for such work to be refined. First, we propose that the relative amounts of dense and non-dense tissues should be considered as two separate outcomes, with both routinely reported. This has been previously suggested [5], but not adhered to in the majority of published studies. This will allow better modelling of associations between risk factors and density, without the potential confounding influence of measures of body fat on the results. We have found that BMI is positively related to SMF volume [21], as would be expected if SMF volume is thought of as a proxy marker for breast cancer risk.

Second, investigations into the biological mechanisms underlying associations between breast density and breast cancer risk are needed. Some work in this area has begun. For example, insulin-like growth factor (IGF) and its main binding protein IGFBP-3 have both been related to breast density [22, 23].

In summary, our findings suggest that the novel technique of estimating the volume of dense breast tissue, which involves modelling mammographic breast density using a fully automated system, has the potential to be a marker of breast cancer risk available for use in large epidemiological studies. Work is underway to investigate whether SMF can predict breast cancer risk.


   Acknowledgments
 
We are grateful to the women who participated in this study for allowing access to their mammograms, to Pat Forrest who performed the digitization and for the assistance of all the Scottish Breast Screening Centres. This work was undertaken while Dr Jeffreys (née Okasha) was employed at the University of Bristol. Dr Highnam is employed by Mirada Solutions, where he developed the SMF algorithm. We are grateful for the financial support provided by Breast Cancer Campaign, Breast Cancer Research Trust and World Cancer Research Fund International. The Centre for Public Health Research is supported by a programme grant from the Health Research Council of New Zealand.

Received for publication April 6, 2005. Revision received July 16, 2005. Accepted for publication September 1, 2005.


   References
Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 

  1. Boyd N, Lockwood G, Byng J, Tritchler D, Yaffe M. Mammographic densities and breast cancer risk. Cancer Epidemiol Biomarkers Prev 1998;7:1133–44.[Abstract/Free Full Text]
  2. Heine JJ, Malhotra P. Mammographic tissue, breast cancer risk, serial image analysis, and digital mammography. Part 1. Tissue and related risk factors. Acad Radiol 2002;9:298–316.[CrossRef][Medline]
  3. Wolfe JN. Breast patterns as an index of risk for developing breast cancer. AJR Am J Roentgenol 1976;126:1130–7.[Abstract]
  4. Gram I, Funkhouser E, Tabar L. Anthropometric indices in relation to mammographic patterns among peri-menopausal women. Int J Cancer 1997;73:323–6.[CrossRef][Medline]
  5. Boyd NF, Lockwood GA, Byng JW, Little LE, Yaffe MJ, Tritchler DL. The relationship of anthropometric measures to radiological features of the breast in premenopausal women. Br J Cancer 1998;78:1233–8.[Medline]
  6. Vachon CM, Kuni CC, Anderson K, Anderson VE, Sellers TA. Association of mammographically defined percent breast density with epidemiologic risk factors for breast cancer (United States). Cancer Causes Control 2000;11:653–62.[CrossRef][Medline]
  7. El-Bastawissi AY, White E, Mandelson MT, Taplin SH. Reproductive and hormonal factors associated with mammographic breast density by age (United States). Cancer Causes Control 2000;11:955–63.[CrossRef][Medline]
  8. Salminen T, Hakama M, Heikkila M, Saarenmaa I. Favorable change in mammographic parenchymal patterns and breast cancer risk factors. Int J Cancer 1998;78:410–4.[CrossRef][Medline]
  9. Highnam R, Brady M. Mammographic image analysis, 1 edn. Kluwer Academic Publishers, 1999.
  10. Marias K, Behrenbruch C, Highnam R, Parbhoo S, Seifalian A, Brady M. A mammographic image analysis method to detect and measure changes in breast density. Eur J Radiol 2004;52:276–82.[Medline]
  11. Hufton A, Astley S, Marchant T, Patel H. A method for the quantification of dense breast tissue from digitised mammograms. Proceedings of the IWDM 2004. 2005 (In press)
  12. Pawluczyk O, Augustine BJ, Yaffe MJ, Rico D, Yang J, Mawdsley GE, et al. A volumetric method for estimation of breast density on digitized screen-film mammograms. Med Phys 2003;30:352–64.[CrossRef][Medline]
  13. McCarron P, Davey Smith G, Okasha M, McEwen J. Life course exposure and later disease: a follow-up study based on medical examinations carried out in Glasgow University (1948–68). Public Health 1999;113:265–71.[CrossRef][Medline]
  14. Jeffreys M, Warren RM, Davey Smith G, Gunnell D. Breast density: agreement of measures from film and digital image. Br J Radiol 2003;76:561–3.[Abstract/Free Full Text]
  15. Byng J, Boyd N, Fishell E, Jong R, Yaffe M. The quantitative analysis of mammographic densities. Phys Med Biol 1994;39:1629–38.[CrossRef][Medline]
  16. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307–10.[CrossRef][Medline]
  17. Wei J, Chan HP, Helvie MA, Roubidoux MA, Sahiner B, Hadjiiski LM, et al. Correlation between mammographic density and volumetric fibroglandular tissue estimated on breast MR images. Med Phys 2004;31:933–42.[CrossRef][Medline]
  18. Silman A, Macfarlane G. Epidemiological studies: a practical guide, 2 edn. Cambridge: Cambridge University Press, 2002.
  19. Byng JW, Boyd NF, Little L, Lockwood G, Fishell E, Jong RA, et al. Symmetry of projection in the quantitative analysis of mammographic images. Eur J Cancer Prev 1996;5:319–27.[CrossRef][Medline]
  20. Vacek PM, Geller BM. A prospective study of breast cancer risk using routine mammographic breast density measurements. Cancer Epidemiol Biomarkers Prev 2004;13:715–22.[Abstract/Free Full Text]
  21. Jeffreys M, Warren R, Gunnell D, McCarron P, Highnam R, Davey Smith G. Body mass index in young adulthood and breast cancer risk (abstract). Australasian Epidemiologist 2003;10:17
  22. Guo YP, Martin LJ, Hanna W, Banerjee D, Miller N, Fishell E, et al. Growth factors and stromal matrix proteins associated with mammographic densities. Cancer Epidemiol Biomarkers Prev 2001;10:243–8.[Abstract/Free Full Text]
  23. Maskarinec G, Williams AE, Kaaks R. A cross-sectional investigation of breast density and insulin-like growth factor I. Int J Cancer 2003;107:991–6.[CrossRef][Medline]



This article has been cited by other articles:


Home page
 RadiologyHome page
P. R. Bakic, A.-K. Carton, D. Kontos, C. Zhang, A. B. Troxel, and A. D. A. Maidment
Breast Percent Density: Estimation on Digital Mammograms and Central Tomosynthesis Projections
Radiology, July 1, 2009; 252(1): 40 - 49.
[Abstract] [Full Text] [PDF]

Home page
 Cancer Epidemiol. Biomarkers Prev.Home page
J. Ding, R. Warren, I. Warsi, N. Day, D. Thompson, M. Brady, C. Tromans, R. Highnam, and D. Easton
Evaluating the Effectiveness of Using Standard Mammogram Form to Predict Breast Cancer Risk: Case-Control Study
Cancer Epidemiol. Biomarkers Prev., May 1, 2008; 17(5): 1074 - 1081.
[Abstract] [Full Text] [PDF]

Home page
 Cancer Epidemiol. Biomarkers Prev.Home page
V. A. McCormack, R. Highnam, N. Perry, and I. dos Santos Silva
Comparison of a New and Existing Method of Mammographic Density Measurement: Intramethod Reliability and Associations with Known Risk Factors
Cancer Epidemiol. Biomarkers Prev., June 1, 2007; 16(6): 1148 - 1154.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via HighWire
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Jeffreys, M
Right arrow Articles by Davey Smith, G
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Jeffreys, M
Right arrow Articles by Davey Smith, G

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
BJR DMFR IMAGING  ALL BIR JOURNALS