Mammographic Breast Density: Effect on Imaging and Breast Cancer Risk

Medscape: Continuing Medical Education Online

Breast Density Measurement

Recognition of breast density as a breast cancer risk factor and measurement of density have been evolving for more than 30 years. This affects a large number of women. More than half of women between 25 and 49 years of age have dense breasts (> 50% density), as do approximately 29% of women older than 50 years.¹ The pioneer of breast density research was John Wolfe, whose early qualitative assessment of breast density describes 4 groups: “N1” as mostly fat, “P1” (less than one fourth of the breast) and “P2” (more than one fourth of the breast) as increasing connective tissue hyperplasia surrounding the ducts, and “DY” as marked mammary dysplasia, which he used to describe the densities. In his early retrospective study, Wolfe reported a 37 times greater incidence of cancer in the “DY” group than the “N1” group, and 82% of the cancers occurred in only 33% of the population representing the “P2” and “DY” groups.² This qualitative (descriptive) assessment of breast density is similar to the now widely used Breast Imaging Reporting and Data System (BI-RADS) of the American College of Radiology. In the BI-RADS system, 4 descriptive categories of breast density are: 1) almost entirely fat, 2) scattered fibroglandular densities, 3) heterogeneously dense, and 4) extremely dense (Figure 1). In the most recent version, BI-RADS density categories became more quantitative, with corresponding quartile assessments of dense areas on the film as 1) less than 25%, 2) 25% to 50%, 3) 51% to 75%, and 4) more than 75%.³

The qualitative assessment of mammographic density has been the basis of many studies evaluating the importance of breast density on both mammographic sensitivity and breast cancer risk.^4,5 Martin et al.⁶ correlated the radiologist's estimate of breast density using BI-RADS qualitative and quantitative scales and a reference standard generated by the physician using manual computerized segmentation of the mammogram into dense and nondense areas. Radiologists not specifically trained in assessing breast density overestimated density by 37% using qualitative scales. A weak correlation was seen between their qualitative BI-RADS breast density assessment and the quantitative percent density, especially for BI-RADS 3 and 4 categories, with estimated percent density ranges of 8% to 60% and 20% to 82%, respectively.

The concern for the inconsistencies in the subjective qualitative estimate of density and the ongoing desire to confirm and quantify breast cancer risk factors have led to the development and testing of computer model assessments of mammographic breast density. All methods, even computer models, remain estimates of “density.” An early method was simple planimetry described by Wolfe in 1976, in which the outline of the breast and the outline of the dysplastic tissue was hand drawn on acetate and then assessed using a planimeter. He described this process as “tedious and time consuming” but not difficult.² The correlation with his qualitative assessment of breast density was good (κ = .91).

More recent methods are partially automated and are less work-intensive. The interactive thresholding technique introduced by Byng et al.⁷ was performed on film screen mammograms that had been digitized or, more recently, on digital mammogram images. The operator must choose a threshold grey value for distinguishing the edge of the breast and a threshold separating dense breast tissue from nondense tissue. The images are then analyzed by computer algorithms that identify the pixels in each category and plot a histogram. The percent density is the area of the dense tissue over the area of the whole breast multiplied by 100. The threshold assignment of dense and nondense areas by grey value is recognized to be a weak point of the method, but it correlates well with qualitative categories and breast cancer risk. This method has been used in many studies to quantify breast density.^8–12 A commercial product is available.¹²

Representations of the 4 Breast Imaging Reporting and Data System (BI-RADS) breast density qualitative and quantitative assessments. A) BI-RADS 1: almost entirely fat; B) BI-RADS 2: scattered fibroglandular densities; C) BI-RADS 3: heterogenously dense; and D) BI-RADS 4: extremely dense.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 8, 10; 10.6004/jnccn.2010.0085

View Full Size

Representations of the 4 Breast Imaging Reporting and Data System (BI-RADS) breast density qualitative and quantitative assessments. A) BI-RADS 1: almost entirely fat; B) BI-RADS 2: scattered fibroglandular densities; C) BI-RADS 3: heterogenously dense; and D) BI-RADS 4: extremely dense.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 8, 10; 10.6004/jnccn.2010.0085

Representations of the 4 Breast Imaging Reporting and Data System (BI-RADS) breast density qualitative and quantitative assessments. A) BI-RADS 1: almost entirely fat; B) BI-RADS 2: scattered fibroglandular densities; C) BI-RADS 3: heterogenously dense; and D) BI-RADS 4: extremely dense.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 8, 10; 10.6004/jnccn.2010.0085

Martin et al.⁶ developed and tested a fully automated computer program for breast density assessment that compared well with radiologists' quantitative analysis and was more reproducible than the qualitative BI-RADS. In this method, computer-applied algorithms perform the periphery detection. Subsequently, a grey-level threshold is applied to separate the dense and nondense areas (Figure 2). The computer-derived mammographic density estimate was then compared with a radiologist computer-generated reference standard, qualitative BI-RADS, and a subjective quantitative 10% scale assessment of density. The fully automated program results correlated highly with the density reference standard. It did not overestimate density as strongly as BI-RADS and the radiologist's subjective percentile quantity assessment.

Although most studies report the PMD as their indicator, absolute breast density is also considered a risk factor for breast cancer. Ursin et al.¹³ proposes that it may be a better predictor of the actual tissue at risk for transformation. Kopans¹⁴ stated that 2-dimensional assessment of breast density, as performed in most studies, is not an accurate representation, calling into question the validity of breast density research to date. Experts have proposed 3-dimensional volumetric analysis as a better reflection of true density. However, 2 independent studies of different computerized methods of volumetric assessment showed that 3-dimensional density measurement did not outperform 2-dimensional for risk prediction.^15,16 The hope is that a rapid and reproducible method of determining breast density will facilitate both patient risk evaluation and breast density research.

Clinical Importance of Breast Density

The importance of mammographic density is primarily 2-fold. It significantly impacts both mammographic interpretation and a woman's personal risk of breast cancer. For those who interpret mammograms and those who use the results for clinical management, it is critical to understand that the sensitivity of mammography for detecting cancer is lower in dense breasts.^17,18 Breast cancer appears white on a mammogram, whether it is a mass, calcifications, distortion, or developing density. The possibility of a cancer being masked by overlying breast tissue is greater in dense breasts than fatty ones.

Computerized measurement of breast density. A) Craniocaudal mammogram; and B) corresponding automated segmented image for computer density assessment. Dense tissue appears white.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 8, 10; 10.6004/jnccn.2010.0085

View Full Size

Computerized measurement of breast density. A) Craniocaudal mammogram; and B) corresponding automated segmented image for computer density assessment. Dense tissue appears white.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 8, 10; 10.6004/jnccn.2010.0085

Computerized measurement of breast density. A) Craniocaudal mammogram; and B) corresponding automated segmented image for computer density assessment. Dense tissue appears white.

Citation: Journal of the National Comprehensive Cancer Network J Natl Compr Canc Netw 8, 10; 10.6004/jnccn.2010.0085

In a large study, Carney et al.¹⁹ studied the individual and combined effects of breast density and other factors on the accuracy of screening mammography in 329,495 women from 7 mammography registries. They described a significantly lower sensitivity of screening mammography in women with extremely dense breasts than in those with almost entirely fatty breasts (62.2% vs. 88.2%, respectively; P < .001). Sensitivity was intermediate for the scattered-density and heterogeneously dense groups (82.1% and 68.9%, respectively). In this study, 7.8% of women had breasts that were classified as extremely dense.¹⁹ Density distribution in the large Digital Mammography Imaging Screening Trial (DMIST) involving approximately 49,000 eligible women also described extremely dense breasts in 8% of their population. Most women had breast densities in the scattered-density or heterogeneously dense categories (43% and 39%, respectively), whereas 10% were in the fatty range.²⁰

Mammographic density has been shown to be a significant independent risk factor for breast cancer dating back to the earliest studies by Wolfe.^2,21 Beyond the dramatic 37-times increased risk described by Wolfe in 1976, multiple studies have shown a consistently increased risk for breast cancer in women with greater breast density. A meta-analysis of several breast density studies to date in 2006 showed a 4- to 6-fold greater risk of cancer in women with the most dense breasts (> 75%) compared with those with the least dense (< 5%).²² A study of 1112 matched case control pairs showed similar results.²¹ This study also showed that the elevated risk persisted for 8 years after the first density measurement.

Mammographic breast density is not a fixed personal characteristic like a fingerprint but rather is a dynamic mammographic representation of the relative amount of fibrous and glandular tissue in the breast relative to fat. Fibrous and glandular tissue cannot be distinguished on the mammogram because both appear white. Density may change over a woman's lifetime under different influences. Chronologic age as a modifier of breast density has been proven in many studies.¹⁹

Kelemen et al.⁹ reviewed 5698 mammograms in 1689 women in a longitudinal study from 1990 to 2003 analyzing percent density with respect to age, menopausal status, postmenopausal hormone use, and body mass index (BMI). Premenopausal women had a higher percent density than postmenopausal women. They showed that with increasing age, breast percent density declined. Over an average range of mammographic follow-up of 6 years, the decrease in density was greatest for the group in transition from premenopause to menopause, with an 8% decline— at least 2-fold greater than in women who remained pre- or postmenopausal during the follow-up period. The decrease in percent density with increasing age seems to contradict the fact that breast cancer incidence increases with age.

Several authors have suggested that the model for breast aging that Pike et al.²³ proposed in 1983 explains this dichotomy. Pike proposed that a cumulative effect of hormones and growth factors on the breast tissue increases the risk of breast cancer. Life factors such as menarche, age at first full-term birth, and timing of menopause change the effective age of the breast and the risk for breast cancer. According to this theory, assessment of breast density at a single point in time would not correlate with risk.

Vachon et al.²⁴ measured the breast density on serial mammograms in the 10 years before a diagnosis of breast cancer and found no association between a change in breast density over time and an increased risk of cancer. This population was 84% postmenopausal, and therefore the concern is that this group may be too advanced in age to show significance in changing breast density.

Maskarinec et al.²⁵ performed a longitudinal study of density over more than 20 years, proposing that cumulative density over time imparts risk, not density at any one point in time. This was explained by the parallel curves of age-specific breast cancer incidence and cumulative breast density. The data showed that factors associated with increased risk of cancer, such as hormone replacement therapy (HRT) and increased postmenopausal BMI, slowed down the expected age-related decrease in breast density and thus increased the cumulative effect.²⁵ The issue of exactly how and when mammographic breast density is linked to the associated increased risk is an area of active research.

Combined estrogen and progestin HRT use has been shown to increase mammographic density and breast cancer risk.^26,27 Breast imaging literature has shown that HRT results in a reversible cause of increasing breast density. Studies have examined the effect of HRT on mammographic sensitivity. Carney et al.¹⁹ concluded that after adjustment for increased breast density, HRT did not have an independent effect on decreasing mammographic sensitivity but rather is related to the effect on breast density.

Analysis of BMI and breast density in several studies has shown the inverse relationship of breast density and BMI. Aitken et al.²⁸ showed that higher socioeconomic status was associated with lower BMI and higher mammographic density.²⁸ Ursin et al.¹³ found that BMI was one of the environmental factors affecting density that blunted the proposed effect of genetic determination. As BMI increases, the amount of fat increases relative to the fibrous and glandular tissue. In the setting of obesity, absolute measurement of density may be more significant.

The heritability of breast density has been examined in several classic twin studies. Both Boyd et al.²⁹ and Ursin et al.¹³ showed that genetic factors contributed to 50% to 60% of the variations in breast density. In North American and Australian twins, Boyd showed correlation coefficients for density of 0.67 and 0.61 in monozygotic pairs and 0.27 and 0.25 in dizygotic pairs, respectively.²⁹ Ursin's results were in the lower range, and the relative contribution of genetics was much lower if greater differences in environmental or modifiable factors were present, such as BMI (as low as 20% contribution) or parity.¹³ A study of breast density in the Old Order Amish population in Lancaster County, Pennsylvania, showed a significant genetic component to the variance of both dense (39%; P < .0001) and nondense (71%; P < 10^–14) areas on mammography.¹¹ A significant genetic component to other breast cancer risk factors was also described, such as parity, age at menarche and menopause, height, weight, and BMI.

Individualized Screening

As science progresses in deciphering the genetic and environmental causes of breast cancer and further elucidates the risk factors, individualized screening is a reasonable goal. The DMIST compared the accuracy of digital mammography with film screen technique in nearly 43,000 women who had both film screen and digital mammograms. In pre- or perimenopausal women younger than 50 years with dense breasts, digital mammography was shown to be more sensitive in cancer detection than film screen technique, with sensitivities of 0.59 versus 0.27, respectively (P < .0013).⁴ Currently, approximately 60% of centers have switched to digital breast imaging.

Importantly, digital mammograms undergo software manipulation of the raw digital data to generate the clinical digital image, unlike film screen, which is unprocessed. Many centers do not store the raw data. Generally, breast density is perceived as somewhat lower on processed digital images than on film screen. This may be beneficial for cancer detection but may make comparison between film screen and digital density less accurate.

No reduction in mortality has been proven for any supplemental screening method other than mammography.³⁰ Individualized screening beyond annual mammography is believed to be most relevant in women at high risk for breast cancer. Currently, women are considered to be high risk if they tested positive for BRCA mutations, are untested first-degree relatives of BRCA carriers, have a history of prior thoracic radiation, or have a lifetime risk for breast cancer of greater than 20% based on cancer risk prediction models, such as BRCAPRO, BOADICEA, and Tyler-Cusick, and a few rare heritable syndromes. These models focus primarily on family history and personal breast history. Breast density is not included in these models. The Breast Cancer Prevention Collaborative Group convened in 2007 in recognition of the limitations of the current risk model.³¹ They have proposed a new comprehensive risk-predicting model that would include breast density, estrogen, and androgen assays and measurements such as BMI.

MRI, with its high sensitivity for detecting invasive cancer, is currently recommended as an annual supplement to mammography in high-risk women. Mammographic breast density does not necessarily correlate with background enhancement with gadolinium on MRI.³² MRI sensitivity does not seem to be as affected by density as is mammography. If mammographic density is incorporated into a clinically useful risk prediction model, then more women may be considered in the high-risk group and would be eligible for supplemental screening. Currently, dense breasts alone are not an indication for MRI screening.

Screening whole-breast ultrasound (WB-US) as a supplement to mammography has been investigated in high-risk women with dense breasts. WB-US with annual mammography has been shown to increase sensitivity in high-risk women with dense breasts compared with mammography alone, but does not seem to equal the added benefit of MRI in the same group.³⁰ Kolb et al.³³ added screening ultrasound to mammography in the assessment of 5418 women of normal and high risk with BI-RADS breast densities of categories 2, 3, or 4. Among the study population, 30% were at increased risk based on having a personal history of breast cancer, a first-degree relative with breast cancer, or a previous high-risk biopsy result. Screening ultrasound was shown to have a higher sensitivity for detecting cancer than mammography (75% vs. 64%, respectively). Ultrasound alone detected 42% of cancers in dense breasts that were not detected using other screening methods.³³

The American College of Radiology Imaging Network (ACRIN) 6666 study of screening breast ultrasound reported improved diagnostic sensitivity when ultrasound was added to screening mammography in high-risk women with dense breasts. This multicenter trial studied 2637 women using mammography alone or with a screening WB-US. The WB-US significantly increased the sensitivity from 50% with mammography alone to 77.5%.³⁴ The downside was an increased false-positive rate and lower positive predictive value (PPV) for a biopsy recommended based on WB-US (PPV: 9% in ultrasound/mammogram group vs. 23% for mammogram alone).³⁴

Each of these WB-US studies has included patients at high risk, and therefore the contribution of density alone is difficult to determine. Screening WB-US for dense breasts has not been recommended by the American Cancer Society, NCCN, or the U.S. Preventative Services Task Force.

Risk Modification

Studies are lacking that prove that modifying breast density changes risk. Tamoxifen has been proven to decrease breast cancer risk and cause decreased breast density in many women.³⁵ Making the mammogram less dense may also diminish the effect of masking. The antiestrogen effect of tamoxifen in the breast may result in involution of the lobules. Stimulation of the growth of the lobules has been proposed as a link between density and cancer risk, with more cells being available for damage by mutagens.³⁶ Modification of other risk factors for breast cancer also can modify density, such as parity, hormone use, and BMI. Beyond showing the common link, studies have not fully elucidated when or how lowering breast density will alter its inherent risk.

Current and future research in breast density includes volumetric quantification of density as a potentially more accurate representation of the 3-dimensional breast¹⁴; development of a reliable, accurate quantification program for assessing mammographic breast density quickly in the clinic; and assessment of breast density pattern variations instead of overall summation of density.

Conclusions

Mammographic breast density is a radiographic representation of the dense fibrous and glandular tissue in the breast relative to fat. Breast density has a proven impact on mammographic interpretation, with dense breasts resulting in a decreased sensitivity for detecting cancer compared with fatty breasts. Density, however imprecisely measured, has been shown to be a significant independent risk factor for breast cancer but is not included in most current risk assessment models. Both subjective and quantitative measures of density are used. Volumetric and fully automated density assessments are areas of active research. Genetic and environmental factors influence breast density. Young women with dense breasts may benefit from the higher cancer detection of digital mammography. When mammographic density is included in clinical risk models, more women may be eligible for supplemental screening methods, such as screening MRI and WB-US, and for genetic counseling and risk reduction efforts.