Study population
Volumetric breast density from 150 CEM examinations were compared with digital mammography or tomosynthesis performed either before or after CEM. On average, prior DM/DBT were obtained 10.8 months before or after the CEM examination; median time interval was 12 months, 91% of cases (134/150) had mammography or tomosynthesis ±15 months before/after CEM.
Table 1 shows characteristics of study population, including age, menopausal status, risk category, and breast density category.
Table 1
Characteristics of study population
Characteristic | Values |
Number of women | 150 |
Age (y) mean±SD median range | 51.0±8.8 51 (35, 76) |
Menopausal status Premenopausal Perimenopausal Postmenopausal | 68 (45.33%) 14 (9.33%) 68 (45.33%) |
Risk category BRCA1 BRCA2 HIGH (lifetime risk > 30%) INTERMEDIATE (lifetime risk > 17% & ≤30%) | 30 (20.00%) 36 (24.00%) 62 (41.33%) 22 (14.67%) |
BIRADS breast density category Predominantly fatty (A) Scattered fibroglandular (B) Heterogeneously dense (C) Extremely dense (D) | 9 (6.00%) 27 (18.00%) 55 (36.67%) 59 (39.33%) |
Background parenchymal enhancement (BPE) Minimal Mild Moderate Marked | 65 (43.33%) 44 (29.33%) 38 (25.33%) 3 (2.00%) |
Previous/subsequent exam Mammography Tomosynthesis | 120 (80.00%) 30 (20.00%) |
BI-RADS Breast Imaging-Reporting and Data System, BRCA1/BRCA2 women with a mutation of the BRCA1 or BRCA2 genes, SD standard deviation
Mean age was 51.0±8.8 years, ranging between 35 and 76 years. Pre-menopausal women were 45.33% (68/150) of enrolled women, while remaining 54.67% (82/150) were either peri- (14/150 = 9.33%) or post-menopausal (68/150 = 45.33%). Most of subjects enrolled in the study were high-risk women (128/150 = 85.33%); 44.00% (66/150) had proven BRCA1 or BRCA2 mutation, and 41.33% (62/150) had family history of breast cancer that, together with other risk factors, including also VBD, lead to lifetime risk for breast cancer above 30%. The remaining 14.67% of women (22/150) had intermediate risk (lifetime risk between 17% and 30%). Women included in the study population had mostly dense breasts: 76.00% (114/150) classified BIRADS C or D, and 24.00% (36/150) classified BIRADS A or B. Regarding the parenchymal background enhancement evaluated with CEM, 72.67% of women showed minimal or mild BPE (109/150), and 27.3% moderate or marked (41/150). Previous or subsequent exams compared to CEM were 80% (120/150) DM and 20% (30/150) DBT.
Volumetric breast density
Table 2 shows results from the Wilcoxon paired test for VBD and all the other variables considered measured from LE-CEM and DM/DBT paired views.
Table 2
Comparison between CEM and DM/DBT of volumetric breast density and related variables.
Variable | LE-CEM median | DM/DBT median | Hodges-Lehmann median difference | 95% confidence interval | P-value |
VBD (%) | 12.73 | 12.39 | 0.075 | -0.19 to 0.34 | 0.5855 |
Breast volume (cm3) | 508.15 | 534.57 | 25.58 | 18.89 to 32.98 | < 0.0001 |
Glandular volume (cm3) | 55.90 | 59.44 | 3.315 | 1.975 to 4.660 | < 0.0001 |
Breast thickness (mm) | 47.7 | 50.0 | 2.35 | 1.90 to 2.80 | < 0.0001 |
Compression force (N) | 102 | 85 | -15.0 | -17.5 to -12.0 | < 0.0001 |
Compression pressure (kPa) | 11.83 | 10.46 | 1.655 | -1.995 to 1.310 | < 0.0001 |
Contact area (mm2) | 8529.46 | 8311.63 | -144.14 | -222.61 to -68.77 | 0.0003 |
Nipple distance from posterior edge (mm) | 88.40 | 88.80 | 0.60 | 0.10 to 1.15 | 0.0180 |
Mean glandular dose (mGy) | 1.544 | 1.527 | -0.043 | -0.074 to -0.012 | 0.0083 |
Median values obtained from LE-CEM and DM/DBT paired views, Hodges-Lehmann median difference and 95% confidence interval, and P-value from the Wilcoxon paired test for VBD and all available variables actually or potentially associated to VBD calculation.
CEM contrast-enhanced mammography, DM digital mammography, DBT digital breast tomosynthesis, VBD volumetric breast density
Median VBD was comparable for LE-CEM and previous/subsequent DM/DBT (12.73% vs. 12.39%), not evidencing any statistically significant difference (P = 0.5855). Conversely, median differences between any other variable pairs were statistically significant (P < 0.05).
Figure 2 shows the correlation plot (on the left) and the Bland-Altman plot (on the right) for VBD measured from LE-CEM and previous/subsequent DM/DBT exams, for single views (upper plots) and for individual cases obtained averaging VBD values from multiple views (bottom plots).
Figure 2
Figure 2: (a) Correlation plot of VBD measured in paired views obtained from LE-CESM and previous/subsequent DM/DBT exams; the Pearson’s correlation coefficient was r = 0.87, a reduction of correlation can be noticed as VBD increases, especially for MLO views. (b) Bland-Altman plot of VBD difference between LE-CESM and previous/subsequent DM/DBT for each paired view; the mean difference was zero, the limits of agreement ±8%. (c) Correlation plot of mean VBD obtained by averaging single-view VBDs for each paired study obtained with LE-CESM and DM/DBT; the Pearson’s correlation coefficient was r = 0.92, the heat map shows that VBD values were mostly grouped below 15%. (d) Bland-Altman plot of VBD difference between LE-CESM and previous/subsequent DM/DBT for each paired study; the mean difference was confirmed to be very close to zero, the limits of agreement reduced to ±5%.
The correlation was good for both per-view and per-case comparison (r = 0.87 and r = 0.92, respectively). In per-case analysis the correlation was slightly improved compared to per-view analysis because some VBD differences for specific views were attenuated by the averaging across multiple views belonging to the same study. The same effect was confirmed by the Bland-Altman plot: in both per-view and per-case plots the mean VBD difference is zero, but the limits of agreement were narrower in per-case than in per-view analysis (about ±6% against ±8%). From per-view plots it can be noticed that the largest VBD differences between LE-CEM and previous/subsequent DM/DBT images predominantly occurred for MLO views. The heat map in the per-case regression plots shows that volumetric breast density values were mostly below 15%. The VBD difference between LE-CEM and previous/subsequent DM/DBT tends to increase with breast density.
Multiregression analysis
Least squares multiple regression (weighted for variance) using per-case dataset showed that breast density variability between CEM and mammography/tomosynthesis was affected by all variables that one can consider to be reflecting differences in breast positioning, with the exclusion of breast volume. The sample case in Fig. 3 shows in the upper part the LE-CEM images of a woman with large fatty breasts, and in the lower part the prior mammograms acquired 12 months before: there is a large difference in breast volumes between the two examinations (CEM: 2634 cm3; DM: 2058 cm3; difference 576 cm3) associated to large difference in breast positioning (CEM mean nipple-to-posterior-edge distance: 174 mm; DM: 156 mm; 18 mm difference), not producing significant variation in volumetric breast density (CEM: 1.9%; DM: 2.7%). Differently, the second sample case in Fig. 4 shows in the upper part LE-CEM images and in the lower part the subsequent (11.5 months later) mammography of a woman with dense breasts; mean VBD changes from 20.1–30.5% because of a better positioning in the second exam. In subsequent mammography breasts were better positioned, including about 1 cm more (nipple-to-posterior-edge distance) compared to CEM resulting in larger volume of fibroglandular tissue (136 cm3 vs. 64 cm3 with CEM) and larger overall breast volume (450 cm3 vs. 319 cm3 with CEM).
Figure 3
Figure 3: (Upper row) LE-CEM images of a woman with large fatty breasts. VBD: 1.9%; breast volume: 2634 cm3; volume of glandular tissue: 51 cm3; breast thickness: 85 mm; nipple-to-posterior-edge distance: 174 mm. (Lower row) DM images acquired 12 months before the CEM exam. VBD: 2.7%: breast volume: 2058 cm3; volume of glandular tissue: 54 cm3; breast thickness: 76 mm; nipple-to-posterior-edge distance: 155 mm. Despite the visible difference in breast positioning, the large reduction in breast volume due to worse positioning in the DM exam did not produce a significant variation of breast density, being the breasts predominantly fatty.
Figure 4
Figure 4: (Upper row) LE-CEM images of a woman with dense breasts. VBD: 20.1%; breast volume: 319 cm3; volume of glandular tissue: 64 cm3; breast thickness: 32 mm; nipple-to-posterior-edge distance: 86 mm. (Lower row) DM images acquired 11.5 months after the CEM exam. VBD: 30.5%; breast volume: 450 cm3; volume of glandular tissue: 136 cm3; breast thickness: 41 mm; nipple-to-posterior-edge distance: 98 mm. Breast positioning is better in DM exam, including an additional volume of glandular tissue leading to a significant increment in VBD.
As reported in Table 3, a VBD difference between LE-CEM and DM/DBT was associated to significant differences of glandular volume, breast thickness, compression pressure, nipple-to-posterior-edge distance, and to differences of compression force and contact area, while P-values for breast volume and MGD difference were above 0.05. The coefficient of determination (R2) was 0.6023, while the multiple correlation coefficient was 0.7761.
Table 3
Coefficients of the regression equation and P-values resulting from the multiple regression model.
Independent variable difference | Regression Coefficient | P-value |
(Constant) | 0.1741 | |
Breast volume (cm3) | 0.005017 | 0.1311 |
Glandular volume (cm3) | + 0.1174 | < 0.0001 |
Breast thickness (mm) | -0.2748 | < 0.0001 |
Compression force (N) | 0.05008 | 0.0036 |
Compression pressure (kPa) | -0.5031 | 0.0001 |
Contact area (mm2) | -0.0009143 | < 0.0001 |
Nipple-to-posterior-edge distance (mm) | -0.1453 | 0.0011 |
Mean glandular dose (mGy) | -0.1373 | 0.6684 |
VBD difference between LE-CEM and DM/DBT was considered the dependent variable; difference in breast volume, glandular volume, breast thickness, contact area, compression force, compression pressure, nipple-to-posterior-edge distance, and mean glandular dose were included in the model as independent variables.
CEM contrast-enhanced mammography, DM digital mammography, DBT digital breast tomosynthesis, VBD volumetric breast density
Figure 5 provides a scatter plot matrix of the VBD differences and the differences between all the independent variables measured in the paired LE-CESM and DM/DBT cases. A scatter plot matrix is a grid (or matrix) of scatter plots used to visualize bivariate relationships between combinations of variables. Each scatter plot in the matrix visualizes the relationship between a pair of variables, allowing many relationships to be explored in one chart. In the upper diagonal, the distribution (histogram) of each variable difference is represented.
Figure 5
Scatter plot matrix of the VBD differences obtained from CEM and DM/DBT images and their relationship with all the other variable differences considered as independent in the multiple regression model: breast volume, volume of glandular tissue, breast thickness, compression force and pressure, contact area, nipple-to-posterior-edge distance, and mean glandular dose.
The only parameter the VBD difference can be considered more than very weakly correlated to is the difference in volume of glandular tissue (r = 0.545). Other correlations with moderate-to-strong relationships are recognizable between differences in breast volume and compressed breast thickness (r = 0.708), between breast volume differences and nipple-to-posterior-edge distance (r = 0.667), and finally between compression pressure differences and compression force differences (r = 0.880).
The Mann-Whitney test for independent samples applied to the subgroup of minimal or mild BPE compared with the subgroup of moderate or marked BPE did not show any statistically significant difference (P = 0.1197).