Evaluation of the association between quantitative mammographic density and breast cancer occurred in different quadrants

The breast tissue mainly consists of two components: fibroglandular tissue and adipose tissue (fat). Fibroglandular tissue is a mixture of fibrous stroma and epithelial cells that line the ducts of the breast. Breast density measured by mammography (MD) is associated with the amount of fibroglandular tissue. Studies of mammographically dense tissues suggest that density may represent increased epithelial cellular concentration, stromal fibrosis, and epithelial hyperplasia [ 1]. MD has been proven as an independent risk factor for BC [ 2– 9]. Women with dense tissue visible on a mammogram have a cancer risk 1.8 to 6.0 times that of women with little density [ 10]. The biological basis for higher cancer risk associated with increased MD is not fully understood. The cellular, biological, and genetic basis of the association between fibroglandular tissue and cancer risk were investigated in many studies, as described in detail in two review articles [ 11, 12]. MD was influenced by hormones and growth factors, and it was hypothesized that the combined effects of cell proliferation (mitogenesis) and genetic damage by mutagens (mutagenesis) led to the increased cancer risk [ 11]. The stroma composed of extracellular matrix proteins, adipocytes, fibroblasts and immune cells is also known to contribute to the increased cancer risk [ 12]. The strong evidence has led to a substantial effort to incorporate breast density into risk prediction models to improve accuracy [ 13– 19].

A fundamental question that has yet to be answered is whether cancers tend to arise in mammographically dense tissue. Among few studies exploring the question, two studies showed that ductal carcinoma in situ (DCIS) [ 20] and invasive cancer [ 21] occurred overwhelmingly in the mammographically dense areas, suggesting that some aspects of glandular/stromal tissue comprising the dense tissue directly influences the carcinogenic process. Another study, however, found that after accounting for the overall percent density (PD) differences, density in the region was not a significant risk factor associated with the location of subsequently developed cancer [ 3].

Many studies have shown that the upper outer quadrant of the breast is the most frequent site for occurrence of breast cancer [ 22– 24]. A study [ 23] consisting of 746 consecutive breast core biopsies noted 62% of 349 malignant lesions (95% confidence interval 57-67%) arose from the UO quadrant. An adequate explanation for this asymmetric occurrence of breast cancer within the breast has never been established. Since density is a risk factor, it would be very interesting to investigate the relationship between quadrant density and the tumor occurring quadrant location. Although this question has been raised for a long time, there were few publications in this area, possibly because of the lack of a reliable method that can measure quantitative density on mammography, as well as the lack of a standardized method that can divide a breast into four quadrants. The inconsistent results in a few published studies reporting quadrant or local breast density might also due to different methods that were used in the analysis [ 3, 20, 21]. Although many studies have reported the measurements of breast density using a variety of imaging modalities and methods, qualitatively or quantitatively, most studies analyzed the density in the whole breast, but not in well-defined quadrants.

In this work we applied a computer algorithm-based segmentation method to quantitatively analyze breast density on mammography, and also applied an established method to divide a breast into 4 quadrants based on craniocaudal (CC) and mediolateral oblique (MLO) mammography using the nipple and the chest wall muscle as references. A breast was separated into: upper-outer (UO), upper-inner (UI), lower-outer (LO), and lower-inner (LI) quadrants; and breast area (BA), dense area (DA) and PD in each quadrant were measured. For each woman, the occurrence of tumor in a specific quadrant was determined, and the women were separated into 4 groups that had tumors in UO, UI, LO, and LI quadrants, respectively. The presence of tumor would affect the measured density; therefore in this study we analyzed the quadrant density of the contralateral normal breast. Despite the fact that some degree of breast asymmetry was expected, the bilateral breasts were considered as symmetric in general, and the normal breast could be used to simulate the diseased breast before the tumor occurred [ 25– 27]. After the women were separated into 4 groups based on tumor location, the BA, DA, and PD of 4 quadrants in the normal breasts of women in these 4 groups were compared to investigate the association of quadrant density with tumor location.

Although MD is associated with breast cancer risk, it is not known whether MD is directly related to cancer occurrence, i.e., tumors arising within the radiodense tissue [ 1]. Higher MD has, histologically, a greater cellular concentration and/or proliferation of the stroma or epithelium [ 20]. It was thus postulated that areas of higher density may be more susceptible to the initiation and promotion of breast cancers than areas of lower density [ 3]. Greater understanding of the association between density and cancer risk may provide information to improve the accuracy of cancer risk prediction and the clinical management of high-risk women [ 3]. Although many studies have investigated and demonstrated that mammographic density was an established risk factor, only a few studies evaluated the association of regional density with the location of the occurred cancer. Two studies measured breast density in different quadrants but showed inconsistent results for the correlation with the occurred cancer [ 3, 20]. Another study applied a computer algorithm to align serial images from the same woman, and used an overlaid grid analysis to measure density in 1-cm squares on prediagnostic mammographic films, and estimated the odds of subsequently developed tumor in relation to its prediagnostic square-specific MD [ 21]. The median prediagnostic MD was 98.2% (46.8%-100%) in 1-cm squares that subsequently contained the tumor, and 41.0% (31.5%-53.9%) in the whole breast [ 21]. The results suggested that tumors were more likely to occur in high MD areas.

In order to perform regional MD analysis, a standardized method to separate a breast into different regions (e.g. quadrants) as well as a reliable segmentation method that can yield quantitative density measurements were needed. In this study we applied a FCM algorithm to perform segmentation and quantify breast density in mammography. Different from Cumulus segmentation, which was focused on the outer boundary, the FCM segmentation was based on the pixel level. FCM could generate consistent results [ 31]; however, the segmentation was highly dependent on the choice of the total cluster number and the clusters that were assigned to differentiate between the dense and fatty tissue. Therefore, we tried to fix the cluster numbers used in the analysis. All final segmentation results were inspected by an experienced radiologist. When the segmentation results were unsatisfactory, the cluster numbers were adjusted. We have also implemented a previously reported method using a bisecting line through the nipple to separate the CC and MLO views into two partitions [ 3, 20]. The breast area and dense tissue area in the UO, UI, LO, and LI quadrants were calculated from the measurements in the CC view medial and lateral regions, and the MLO view superior and inferior regions.

Our datasets were from a cross-sectional study, and the women already had cancer in one breast. To overcome this problem, we analyzed the density in the contralateral normal breast, assuming that it mirrored the density in the diseased breast before the tumor occurred. In most women the bilateral breasts were in general symmetrical [ 25– 27]. A study [ 25] to investigate the spatial distribution of density within the breast using 493 mammographic images from a sample of 165 premenopausal women showed that the degree of the spatial clustering of density was similar between a woman’s two breasts, and did not change with aging. We have also compared the measured density in the quadrants of the bilateral breasts which had no visible tumors, shown in Fig. 7. In each group of women with tumor in a specific quadrant, PD in the three quadrants which had no visible tumor was measured, and the results in the left and right breasts were compared. The Pearson correlation showed a strong correlation coefficient with r = 0.90, suggesting the validity of bilateral symmetry. If the tumor was very big, the presence of tumor might shift the density distribution or affect the separation of a breast into 4 quadrants in the diseased breast. The mean tumor size was 2.1 cm, relatively small within the mean breast area of 116 cm ². Without the presence of tumor, the left-right symmetry was expected to be better. The diseased breast was mainly used to determine the tumor location. The majority of presented results in this study were analyzed from the normal breast.

Given that the premise of the hypothesis in this study depended on bilateral breast symmetry, it would be preferable to analyze each patient’s symmetry utilizing more remote mammograms prior to the detectable cancer. Unfortunately we did not have that dataset in our current study, thus were unable to carry out the analysis. Overall, the assessment of symmetry in mammography is potentially limited by the fact that natural distortions between breasts are likely to occur during the course of breast compression routinely used in mammography. As such, symmetry measures can be confounded by the nature of the imaging procedure itself [ 26]. In our recently published results using 3D MRI in the study of breast density in 58 normal women, 47 pre-menopausal and 11 post-menopausal women [ 32], we found that bilateral breasts in women without cancer are highly symmetrical ( r = 0.97 for breast volume, r = 0.97 for fibroglandular tissue volume, and r = 0.98 for PD). Another study using MRI showed small differences in the bilateral breast tissue composition, i.e. fat and water content, in young women and adults [ 33].

Our results showed that breast cancer was the most likely to occur in the UO quadrant (60.9%). This finding was consistent with most of the published studies in Western women [ 22– 24], Eastern women [ 34], and Asian women [ 35, 36]. A study of Taiwanese women [ 18] showed that more than half (52.3%) of the primary breast tumors occurred in the UO quadrant. Other studies showed that the UO quadrant is also the most frequent location in many benign breast conditions including fibroadenoma and breast cysts [ 37], and phyllodes tumor [ 38]. The reasons why breast cancer occurs more frequently in the UO quadrant are not clear. One study reported that the high proportion of UO quadrant breast carcinomas was a reflection of the greater amount of breast tissue in this quadrant [ 23]. Another study found a disproportional annual increase in breast cancer in the UO quadrant, and that the proportion of UO quadrant breast cancer was the highest in the youngest age group [ 24], and it was postulated that the high rate of UO cancer might be related to the increasing use of cosmetics applied to the adjacent underarm and upper breast area. The underarm cosmetics are known to contain both DNA-damaging chemicals and chemicals which can mimic estrogen action [ 39], and the use of these cosmetics was reported to be associated with younger age for breast cancer diagnosis [ 40]. A recent study of genomic patterns of loss of heterozygosity and allelic imbalance in breast quadrants from 21 breast cancer patients showed increased levels of genomic instability in the outer breast quadrants, suggesting that a higher breast cancer rate in the UO quadrant might result from the development of genetic alterations in that region of the breast rather than from only a greater tissue volume [ 41]. These studies were mainly speculations, and so far there were few mechanistic studies published in the literature to investigate the etiology leading to the higher UO cancer occurrence rate.

Our results showed that among the four breast quadrants, the UO quadrant had the highest mean DA, and thus a larger amount of dense tissue might appear to be associated with the higher cancer rate in the UO. However, if the relationship between the amount of density tissue and cancer occurrence rate is true, the similar finding of higher density and higher cancer rate in the UO should be seen in cancers occurring in the other three quadrants as well. We first compared mean BA and DA among women with tumors occurring in the 4 quadrants. In each group the order was UO > UI > LO > LI. Also, after normalizing DA with BA, the order of the PD in each of the 4 groups was the same: LO > LI > UO > UI. Therefore, the quadrant area and density in these 4 groups of women were very similar, and the statistical analysis results showed that there was no significant difference between them. Then, we applied a “ranking” analysis method to investigate their relationship. For the DA, tumor in UI has the second highest DA in UI, tumor in LO has the third highest DA in LO, and tumor in LI has the lowest DA in LI. The order is exactly the same as the ranking of DA in 4 quadrants from the quadrant separation, i.e. UO > UI > LO > LI. For PD, the cancer-occurring quadrant had random rankings, and no trend at all. Only 30 of 110 women (27.3%) had cancer occurring in the quadrant with the highest PD. Therefore, there was no evidence to support that breast cancer was more likely to occur in the quadrant with the highest DA or PD. Our results concurred with a mammographic study [ 3] that regional breast density was not a significant risk factor for the subsequent development of breast cancer. Another study also concluded that a greater amount of breast tissue in a specific region could not solely explain the preference of breast cancer in the UO quadrant [ 24].

Besides quadrant PD, we also analyzed overall PD in the contralateral normal breast of the 110 women. Although in this study we did not have a matched case control group for the comparison, literature report on the comparison of MD between case and control groups in Asian women has shown significant difference in both pre- and post-menopausal women [ 42]. Two other studies, however, only showed significant density difference of the two groups in the postmenopausal women [ 43, 44].

In conclusion, in this study we used a computer-algorithm based segmentation method to quantify the dense tissue based on mammographic images, and also applied a standardized method to separate a breast into 4 quadrants. We found breast cancer was most likely to occur in the UO quadrant, which was also the quadrant with the highest BA and DA. However, for women with cancers occurring in UI, LO, or LI quadrant, the density in that quadrant was not the highest. When the breast area and density in the four groups of women with tumors occurring in four quadrants were compared, the results were very similar. All 4 groups of women showed the order of BA and DA as UO > UI > LO > LI, and the order of PD as LO > LI > UO > UI. Less than one quarter of women had breast cancer occurring in the quadrant with the highest PD. Our results showed that the differences in quadrant density was mainly from the breast division method, not related to the cancer occurrence. The amount of breast tissue in a specific quadrant cannot explain the preference of breast cancer occurring in a specific location.