Physics in Medicine & Biology

PAPER

Average glandular dose in paired digital mammography and digital

breast tomosynthesis acquisitions in a population based screening
program: effects of measuring breast density, air kerma and beam
quality
To cite this article: Bjørn Helge Østerås et al 2018 Phys. Med. Biol. 63 035006

View the article online for updates and enhancements.

This content was downloaded from IP address 144.96.173.4 on 10/02/2018 at 04:16

 Phys. Med. Biol. 63 (2018) 035006 (14pp) https://doi.org/10.1088/1361-6560/aaa614

PAPER

Average glandular dose in paired digital mammography and digital

breast tomosynthesis acquisitions in a population based screening
RECEIVED
18 May 2017

program: effects of measuring breast density, air kerma and beam

RE VISED
5 December 2017

quality
ACCEP TED FOR PUBLICATION
8 January 2018
PUBLISHED
25 January 2018
Bjørn Helge Østerås1,2, Per Skaane2,3, Randi Gullien3 and Anne Catrine Trægde Martinsen1,4
1
Department of Diagnostic Physics, Oslo University Hospital, Oslo, Norway
2
Institute of Clinical Medicine, University of Oslo, Oslo, Norway
3
Department of Radiology and Nuclear Medicine, Oslo University Hospital, Oslo, Norway
4
Institute of Physics, University of Oslo, Oslo, Norway
E-mail: Bjorn.Helge.Osteras@ous-hf.no

Keywords: mammography, digital breast tomosynthesis, average glandular dose, breast density, screening

Abstract
The main purpose was to compare average glandular dose (AGD) for same-compression digital
mammography (DM) and digital breast tomosynthesis (DBT) acquisitions in a population based
screening program, with and without breast density stratification, as determined by automatically
calculated breast density (Quantra™). Secondary, to compare AGD estimates based on measured
breast density, air kerma and half value layer (HVL) to DICOM metadata based estimates. AGD
was estimated for 3819 women participating in the screening trial. All received craniocaudal
and mediolateral oblique views of each breasts with paired DM and DBT acquisitions. Exposure
parameters were extracted from DICOM metadata. Air kerma and HVL were measured for all
beam qualities used to acquire the mammograms. Volumetric breast density was estimated using
Quantra™. AGD was estimated using the Dance model. AGD reported directly from the DICOM
metadata was also assessed. Mean AGD was 1.74 and 2.10 mGy for DM and DBT, respectively. Mean
DBT/DM AGD ratio was 1.24. For fatty breasts: mean AGD was 1.74 and 2.27 mGy for DM and DBT,
respectively. For dense breasts: mean AGD was 1.73 and 1.79 mGy, for DM and DBT, respectively. For
breasts of similar thickness, dense breasts had higher AGD for DM and similar AGD for DBT. The
DBT/DM dose ratio was substantially lower for dense compared to fatty breasts (1.08 versus 1.33).
The average c-factor was 1.16. Using previously published polynomials to estimate glandularity
from thickness underestimated the c-factor by 5.9% on average. Mean AGD error between estimates
based on measurements (air kerma and HVL) versus DICOM header data was 3.8%, but for one
mammography unit as high as 7.9%. Mean error of using the AGD value reported in the DICOM
header was 10.7 and 13.3%, respectively. Thus, measurement of breast density, radiation dose and
beam quality can substantially affect AGD estimates.

1. Introduction

A major challenge in mammography screening is that the x-ray attenuation of lesions and fibroglandular tissue
is similar. Thus, fibroglandular tissue can obscure lesions in digital mammography (DM), resulting in reduced
lesion detectability and lesion characterization. It has been shown that mammograms of breasts with a large
proportion of fibroglandular tissue (dense breasts) leads to reduced sensitivity compared to fatty breasts (Carney
et al 2003). Digital breast tomosynthesis (DBT) produces an image stack, where a thin slice of anatomy is in focus
in each of the images, while the rest is blurred. This enables the reader to ‘see through’ overlying fibroglandular
tissue, thus potentially reducing the masking effect. DBT has been shown to improve the cancer detection rate in
mammography screening (Skaane et al 2013, 2014), especially in dense breasts (Bernardi et al 2016, Rafferty et al
2016). Therefore, DBT is currently under evaluation as an alternative modality for mammography screening.

© 2018 Institute of Physics and Engineering in Medicine

Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

In addition to the masking effect, breast density has been shown to increase the risk of attaining breast cancer
(McCormack and dos Santos Silva 2006). This has motivated the introduction of commercial software for
automatic breast density calculation, which facilitates objective large-scale density assessment in mammography
screening.
In mammography screening, asymptomatic women are exposed to radiation. Therefore, it is important to
quantify radiation dose. Still, this dose penalty has been estimated to be small compared to the expected benefit
of mammography screening (Yaffe and Mainprize 2011). The dose quantity used for radiation dose assessment
in mammography is average glandular dose (AGD), since the glandular tissue is most radiosensitive in the breast.
Thus, incorporation of breast density in the dosimetry estimates increases the precision of the dose estimates.
In Europe, AGD is primarily estimated using conversion factors based on a model published by Dance (1990)
and Dance et al (2000, 2009, 2011). Here, the breast density is accounted for by the ‘c-factor’. In this study, we
consider two definitions of breast density. The first is volumetric density, defined as the ratio of fibroglandular- to
total breast volume. This is the breast density commonly exported by breast density assessment software, such as
Quantra™ (Hologic Inc., Marlborough, MA, USA) and Volpara™ (VolparaSolutions, Wellington, NZ). The sec-
ond is glandularity, which is the proportion by weight of the fibroglandular tissue in the breast, except for a 5 mm
layer of subcutaneous fat. There have been several publications comparing the radiation dose between DM and
DBT (Tagliafico et al 2013, Bouwman et al 2015, Shin et al 2015, Svahn et al 2015). To the authors knowledge none
of these studies incorporated woman-specific breast density into the model. Studies by Olgar et al (2012) and
Castillo-Garcia et al (2017) estimated AGD in DM and DBT acquisitions incorporating woman-specific breast
density. However, these studies used volumetric breast density, rather than glandularity, which is the density
measure of the model published by Dance et al. In addition, the cohorts were not representative for mammogra-
phy screening as they include referred patients. A recent study by Gennaro et al (2018) compared DM and DBT
doses in the STORM-2 trial population, accounting for glandularity using VolparaDose™ (described below).
However, air kerma and beam quality measurements were not reported in their study. To the authors knowledge,
there is no published results comparing AGD between DM and DBT in a population based screening setting,
using paired DM and DBT acquisitions, which incorporates woman-specific breast density accompanied with
measurements of air kerma and beam quality.
Commercial software estimating and/or reporting AGD automatically has become available. One such
product: VolparaDose™ (VolparaSolutions, Wellington, NZ) estimates radiation dose based on Digital Imag-
ing and Communications in Medicine (DICOM) metadata, breast density and the Dance model (Tromans
et al 2014). Other software such as DoseWatch™ (GE Healthcare, Buc, FR), DoseTrack™ (Sectra, Lindköping,
SWE), tqm|dose™ (Quaelum, Leuven, BE) and Radimetrics™ (Bayer HealthCare LCC, Whippany, NJ, USA) also
designed for other types of radiological modalities, provides AGD estimates. To the authors knowledge these
software use either the AGD value reported in the DICOM metadata or possibly a dosimetry model based on air
kerma or half value layer (HVL) values reported by the DICOM metadata. For such applications, the potential
error of using the DICOM metadata versus using measured air kerma and HVL is of interest. Additionally, if
manufacturers do not measure and include breast density in their dose estimates, it could be accounted for by age
or not accounted for in the AGD estimate, which would contribute to error in the AGD estimates.
The main purpose of this study was to compare AGD in paired DM and DBT acquisitions in a population
based screening program, with and without breast density stratification, as determined by automatically calcu-
lated breast density (Quantra™). The secondary purpose was to analyze effect of incorporating measurements of
breast density in AGD estimates, and to assess the difference in AGD estimates based on DICOM metadata versus
measurements.

2.  Materials and methods

2.1.  Screening trial study population and imaging

The cohort used in this study was drawn from the Oslo Tomosynthesis Screening Trial, where 24313 women went
through screening in the period from November 22, 2010 to December 19, 2012. The study was approved by the
ethical committee, and all women gave informed consent. Imaging was performed using three Hologic Selenia
Dimensions units (Hologic Inc., Marlborough, MA, USA). Craniocaudal (CC) and mediolateral oblique (MLO)
views were acquired of both breasts in a combo mode (DM plus DBT acquisition during the same compression).
All images were acquired using the standard screening automatic exposure control (AEC) setting: ‘auto filter’.
The mammography units use the following target/filter combinations: W/Rh (50 µm) and W/Ag (50 µm) for
DM and W/Al (0.7 mm) for DBT.

2.2.  Population included for dose calculation

Air kerma and HVL of all clinically used radiation qualities were measured in November 2012. As tube output
and beam quality can be adjusted at periodic service, it was assumed that these measurements would be most

2
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

representative within the current service interval. Thus, the dose survey was limited to the 4178 women imaged
between August 6 and November 20, 2012 to ensure the highest quality of data. Images were imported to a
SecurView Workstation (Hologic Inc., Marlborough, MA, USA), where a script (DicomDumpGemini) obtained
from Hologic extracted the DICOM metadata. 278 women had missing data from one or more views. Breast
density data was obtained using Quantra™ version 2.0 (Hologic Inc., Marlborough, MA, USA) (described
below). For 73 women, there was missing density data for one or more views. 8 women were excluded due to
receiving exposures using a combination of Ag filter and 37 kVp. Preliminary investigation in this study did not
report exposures of this beam quality; thus, this spectrum was not measured. The women with missing DICOM
and/or Quantra™ data were excluded, and the final dataset included 3819 women, resulting in 15276 paired DM
and DBT views. This final cohort had a mean age of 58.8 with a range of 50–70 years. The mean compressed
breast thickness was 53.4 mm (sd 12.9, range 14–101).

2.3.  Volumetric density assessment software

Quantra™ version 2.0 was used in this study. Quantra™ estimates the breast density automatically based on the
raw DM image, the DICOM metadata and a physical model of the x-ray imaging chain (Johns and Yaffe 1987,
Boone et al 1997, Hartman et al 2008, Yaffe 2008, Hologic 2012). Quantra™ estimates the height of fibroglandular
tissue which must have been present between each pixel and the compression plate. The algorithm corrects for
the attenuation effects of the skin tissue and excludes the pectoral muscle on MLO views. The fibroglandular
height at each pixel is aggregated into volume fibroglandular tissue. Quantra™ also outputs the total volume of
the breast and the breast density in percent.
Additionally, Quantra™ also reports an ordinal ‘BI-RADS density like’ score from 1 to 4: Quantized density.
Quantized density 1 and 2 is considered fatty and 3 and 4 dense. The cutoffs in volumetric density between each
category are based on a mapping of volumetric density to a reference population (DMIST trial) (Pisano et al
2005, Hologic 2012).

2.4. Measurements
Measurements on all three mammography units for all clinically used kVp and filter combinations were
performed, using a 10  ×  6-6M mammographic ion chamber (RadCal Corporation, Moravia, CA, USA). A thick
steel plate was positioned over the breast support to protect the imaging system and minimize backscatter. Air
kerma was measured with the sensitive area of the detector 4.5 cm above the breast support, centered laterally and
4 cm from the chest edge (IAEA 2007, 2011), with the compression paddle in contact with the dosimeter. HVL was
measured using the ion chamber in the same position as described above. The compression paddle was raised to
about half way between the breast support and the x-ray tube. The beam was collimated using a lead diaphragm
positioned on top of the compression paddle. Aluminum sheets (RadCal Corporation, Moravia, CA, USA) of
at least 99.9% purity with 0.1 mm thickness was inserted above the diaphragm. HVL was calculated according
to the standard formula for HVL calculations (Perry et al 2006, IAEA 2011). The air kerma measurements were
scaled to the square law to reflect air kerma in the entrance plane of the breast. The focus breast support distance
of the mammography units was 67.5 cm.

2.5.  Dosimetry model and glandularity

AGD was estimated using the International Atomic Energy Agency (IAEA) protocol (IAEA 2007, 2011), where
the incident entrance air kerma is used to estimate AGD through several conversion factors (equation (1)).
D = K · g · c · s · T.
(1)
Where D is the AGD, K is the air kerma. g is the g-factor, which is the air kerma to AGD conversion factor. c is the
c-factor which accounts for the glandularity of the breast. s is the s-factor which accounts for x-ray spectrum
differences due to various anode filter combinations. To accommodate DBT acquisitions, the formula is modified
according to Dance et al (2011), by introducing a T-factor, T. These factors are tabulated in Dance (1990), Dance
et al (2000, 2009, 2011) and Perry et al (2013), and were linearly interpolated to a resolution of 1 mm for breast
thickness, 1% for glandularity and 0.001 mm Al for HVL.
In this model (‘Dance model’) the breast modelled as a compressed CC-view, a semi-circle. In this study, the
term ‘glandularity’ was used to describe the breast density definition used by the Dance model (described in the
following paragraph), which should not be confused with volumetric density (exported by Quantra™ or similar
software).
The glandularity refers to the fibroglandular fraction by weight in a central region (core) of the breast. The
reason for this is that the fibroglandular tissue is assumed to be located centrally, surrounded by 5 mm adipose
tissue or all sides except against the chest wall, including the distal semi-circle of the breast (Dance 1990). How-
ever, the glandularity estimates most commonly used today are derived from modelling numerous exposure
factors in the British screening program using tissue equivalent phantoms (Young et al 1996, Becket et al 2000).

3
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

These glandularity estimates were applied by Dance et al (2000) when introducing the c-factor. Unlike the origi-
nal model used by Dance (1990), these glandularity estimates does not account for 5 mm adipose tissue in the
distal semi-circle of the breast. To avoid introducing additional bias between our measurements and the glandu-
larity estimates commonly used today, we chose to only account for 5 mm adipose tissue on the top and bottom
of the breast and not the distal semi-circle. Thus, to select the appropriate c-factor, the volumetric breast density
calculated by Quantra™ was converted to glandularity using equations (2)–(9).
v =a·t
(2)
v
a=
(3)
t

vcore = a · (t − 1)
(4)
v
vcore = · (t − 1)
(5)
t
vf,core = vcore − vg,Q
(6)

wf = vf,core · 0.93 g cm−3

(7)

wg = vg,Q · 1.04 g cm−3

(8)

wg Dg  100%, Dg = Dg
Dg =
(9) , .
wf + wg Dg > 100%, Dg = 100%

Where a is the estimated breast area (assuming constant breast area between from detector to compression
paddle), v is the breast volume exported by Quantra™, and t the breast thickness in cm from the DICOM
metadata. vcore is the estimated breast core volume. vf,core is the estimated fat volume of the breast core. vg,Q is the
fibroglandular volume exported by Quantra™. wf and wg is the mass of fat and glandular tissue in the breast core
using the respective density values (Hammerstein et al 1979). Dg is the glandularity defined in the Dance model.
Tromans et al (2014) used data obtained from Volpara™ to obtain similar estimates of glandularity.

2.6.  Statistical test

Statistical significance was evaluated using a two-sample t-test (ttest2, Matlab) (Mathworks, Natick, MA, USA).
The goodness of fit was evaluated using the coefficient of determination (r2).

3. Result

3.1.  AGD for all women

Table 1 shows the AGD values at compressed breast thickness intervals of 10 mm for DM, DBT and combo
acquisitions, and the DBT/DM dose ratio. The mean AGD per view for all women was 1.74, 2.10 and 3.84 mGy
for DM, DBT and combo acquisitions respectively. The mean dose ratio between DM and DBT acquisitions was
1.24. The DM acquisitions showed a greater variation in dose than DBT within each thickness strata.

3.2.  AGD stratified by breast density

Table 2 shows dose per view for DM, DBT and combo acquisitions, and the DBT/DM dose ratio stratified by
fatty (Quantized density 1 or 2) and dense (Quantized density 3 or 4) breasts. The overall AGD difference is
small for DM: 1.74 versus 1.73 mGy (p  =  0.13) for fatty and dense breasts respectively. For DBT the overall
difference is larger: 2.27 versus 1.79 mGy (p  <  0.001). For combo acquisitions the difference was 4.01 versus 3.51
mGy (p  <  0.001). The difference in DBT/DM dose ratio between fatty and dense breasts was: 1.33 versus 1.08
(p  <  0.001).
For DM, the mean dose within each stratum is higher for dense than fatty breasts (p  <  0.001 within each
stratum). For DBT breasts with thickness  <40 mm have slightly higher (p  <  0.001 within each stratum) DBT
dose to dense breasts compared to fatty. At thickness 40–49 mm the DBT doses for fatty and dense breasts are bor-
derline significantly different (p  =  0.04). For thickness  >50 mm the DBT dose is slightly higher for fatty breasts
(p  <  0.001 within each stratum). The variation in AGD with breast thickness stratified by fatty and dense breasts
is shown for: DM in figures 1(a) and (b), DBT in figures 1(c) and (d), Combo acquisitions in figures 2(a) and
(b) and the DBT DM dose ratio is shown in figures 2(c) and (d). There are about 70 outliers with low dose in the
DM exposures for large breasts, figure 1(a). These outliers are a subset of the images taken of large breasts on one
mammography unit (lab 2). These outliers also affect the combo doses and dose ratios in figures 2(a) and (c).

4
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Table 1.  Radiation dose per view stratified by compressed breast thickness.

AGD (mGy)
a b
CBT (mm) Number of views Mean Median 25–75th percentile

DM 10–19 54 0.83 0.82 0.77–0.89

20–29 573 1.02 1.01 0.94–1.09
30–39 1706 1.17 1.15 1.04–1.28
40–49 3189 1.34 1.29 1.15–1.48
50–59 4759 1.70 1.64 1.47–1.88
60–69 3515 2.22 2.16 1.93–2.44
70–79 1218 2.53 2.48 2.26–2.75
80–89 217 2.54 2.64 2.31–2.90
90–99 40 2.16 1.69 1.50–3.04
All 15276 1.74 1.63 1.29–2.10
DBT 10–19 54 1.35 1.34 1.27–1.41
20–29 573 1.29 1.28 1.21–1.36
30–39 1706 1.33 1.32 1.25–1.40
40–49 3189 1.59 1.58 1.49–1.68
50–59 4759 2.04 2.06 1.90–2.19
60–69 3515 2.64 2.62 2.44–2.81
70–79 1218 3.25 3.26 3.09–3.41
80–89 217 3.72 3.71 3.55–3.86
90–99 40 3.90 3.88 3.84–3.95
All 15276 2.10 2.02 1.58–2.50
COMBO 10–19 54 2.18 2.16 2.06–2.29
20–29 573 2.31 2.29 2.15–2.45
30–39 1706 2.50 2.47 2.29–2.67
40–49 3189 2.92 2.88 2.68–3.10
50–59 4759 3.75 3.73 3.44–4.01
60–69 3515 4.86 4.80 4.44–5.21
70–79 1218 5.78 5.75 5.47–6.04
80–89 217 6.26 6.31 5.97–6.80
90–99 40 6.05 5.47 5.30–6.99
All 15276 3.84 3.69 2.89–4.62
DBT/ DM 10–19 54 1.63 1.60 1.56–1.65
20–29 573 1.27 1.27 1.23–1.31
30–39 1706 1.15 1.13 1.07–1.20
40–49 3189 1.22 1.19 1.07–1.35
50–59 4759 1.24 1.25 1.07–1.40
60–69 3515 1.22 1.24 1.07–1.37
70–79 1218 1.33 1.32 1.18–1.46
80–89 217 1.65 1.42 1.26–1.66
90–99 40 2.20 2.26 1.30–2.63
All 15276 1.24 1.22 1.09–1.37

a
CBT: compressed breast thickness
b
Five views with thickness 100–109 cm were excluded in the CBT stratified analysis due to low number of views in this thickness
interval compared to the number of views already excluded due to lack of spectral measurements at Ag filter with 37 kVp. The five
views are included in the analysis of all thicknesses.

3.3.  Effect of incorporating glandularity into dose estimates

The mean volumetric breast density was 11.2% (median 9, 25th to 75th percentile range of 6–14).
Figure 3(a) shows the estimated glandularity calculated from the breast density measurements from
­Quantra™. The mean and median glandularity was 15.9 and 12%, respectively. With a 25th to 75th percentile
range of 8–25%. The 3rd degree polynomial fit for women between 50 and 64 years published by Dance et al
(2000) is shown along with a 3rd degree polynomial fit to data from this study, shown in equation (10), where t is
the compressed breast thickness. The goodness of fit (r2) was 0.36.
glandularity (%) = −0.000 203 03t 3 + 0.0465t 2 − 3.478t + 100.2.
(10)

5
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Table 2.  Radiation dose per view stratified by compressed breast thickness and breast density.

Fatty breast (Quantized density 1 and 2) Dense breast (Quantized density 3 and 4)

AGD (mGy) AGD (mGy)

CBTa Number 25–75th Number 25–75th

(mm) of viewsb Mean Median percentile of viewsc Mean Median percentile

DM 10–19 0 — — — 54 0.83 0.82 0.77–0.89

20–29 88 0.94 0.94 0.89–0.99 485 1.04 1.03 0.96–1.11
30–39 569 1.03 1.03 0.93–1.10 1137 1.24 1.23 1.13–1.34
40–49 1680 1.19 1.18 1.08–1.29 1509 1.50 1.47 1.32–1.63
50–59 3383 1.58 1.56 1.42–1.73 1376 1.99 1.96 1.74–2.19
60–69 2863 2.12 2.08 1.89–2.31 652 2.68 2.64 2.33–2.98
70–79 1040 2.44 2.43 2.24–2.64 178 3.03 3.02 2.74–3.40
80–89 182 2.42 2.58 2.28–2.81 35 3.18 3.24 2.55–3.73
90–99 35 1.95 1.59 1.43–2.86 5 3.62 3.94 2.99–4.04
All 9844 1.74 1.68 1.32–2.11 5432 1.73 1.55 1.25–2.06

DBT 10–19 0 — — — 54 1.35 1.34 1.27–1.41

20–29 88 1.22 1.21 1.16–1.27 485 1.30 1.30 1.22–1.38
30–39 569 1.26 1.26 1.19–1.33 1137 1.36 1.35 1.28–1.43
40–49 1680 1.58 1.58 1.48–1.68 1509 1.59 1.58 1.49–1.68
50–59 3383 2.08 2.09 1.94–2.22 1376 1.96 1.96 1.82–2.10
60–69 2863 2.66 2.64 2.47–2.83 652 2.52 2.49 2.34–2.69
70–79 1040 3.27 3.27 3.11–3.42 178 3.17 3.16 3.02–3.31
80–89 182 3.74 3.72 3.58–3.93 35 3.57 3.61 3.39–3.73
90–99 35 3.90 3.88 3.84–3.95 5 3.89 3.86 3.75–4.08
All 9844 2.27 2.21 1.78–2.71 5432 1.79 1.64 1.41–2.04

COMBO 10–19 0 — — — 54 2.18 2.16 2.06–2.29

20–29 88 2.15 2.15 2.05–2.25 485 2.34 2.33 2.18–2.49
30–39 569 2.29 2.30 2.14–2.41 1137 2.60 2.58 2.41–2.77
40–49 1680 2.77 2.77 2.60–2.94 1509 3.09 3.05 2.82–3.30
50–59 3383 3.66 3.66 3.39–3.93 1376 3.95 3.90 3.60–4.26
60–69 2863 4.78 4.73 4.40–5.13 652 5.21 5.12 4.75–5.59
70–79 1040 5.71 5.70 5.45–5.98 178 6.20 6.17 5.79–6.58
80–89 182 6.16 6.23 5.83–6.61 35 6.75 6.94 6.22–7.42
90–99 35 5.85 5.44 5.22–6.78 5 7.51 8.01 6.80–8.03
All 9844 4.01 3.92 3.12–4.83 5432 3.51 3.21 2.65–4.11

DBT/ DM 10–19 0 — — — 54 1.63 1.60 1.56–1.65

20–29 88 1.30 1.30 1.26–1.33 485 1.26 1.26 1.22–1.30
30–39 569 1.24 1.21 1.16–1.30 1137 1.10 1.10 1.05–1.15
40–49 1680 1.34 1.34 1.22–1.45 1509 1.08 1.07 1.01–1.14
50–59 3383 1.33 1.33 1.21–1.44 1376 1.01 0.99 0.90–1.09
60–69 2863 1.28 1.28 1.16–1.39 652 0.98 0.95 0.85–1.05
70–79 1040 1.37 1.35 1.23–1.48 178 1.09 1.03 0.92–1.18
80–89 182 1.74 1.47 1.31–1.71 35 1.21 1.10 0.93–1.41
90–99 35 2.36 2.36 1.41–2.68 5 1.10 1.04 0.99–1.26
All 9844 1.33 1.31 1.20–1.43 5432 1.08 1.07 0.97–1.16

a
CBT: compressed breast thickness.
b
Four and c one view with CBT 100–109 cm were excluded in the CBT stratified analysis due to low number of views in this thickness
interval compared to the number of views already excluded due to lack spectral measurements at silver filter with 37 kVp. These four
and one view are included in the analysis of all thicknesses.

Figure 3(b) shows the c-factors used in the dose estimates in this cohort. The average c-factor was 1.16. The
c-factor increased with breast thickness and plateaus for 70–80 mm breasts.
The difference in percent between c-factors derived using glandularity based on measured breast density and
glandularity derived from the Dance polynomial was calculated. The mean difference in c-factor was  −5.9%.
The c-factor difference was largest for thinner breasts with Dance polynomials underestimating the c-factor

6
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Figure 1.  Variation in dose with thickness for DM acquisitions of women with fatty (a) (quantized density 1 or 2) and dense (b)
(quantized density 3 or 4) breasts, and for DBT acquisitions of women with fatty (c) and dense (d) breasts. The colormap represents
the number of views with resolution of 1 mm along the x-axis and 0.06 mGy along the y-axis.

compared to c-factors based on measurements. The value of the c-factor for larger breasts using the Dance poly-
nomials is greater than that obtained by considering the breast glandularity measured in this work.
Figure 3(c) shows the difference in percent between c-factors derived using glandularities based on measure-
ments and glandularities derived from the polynomial given in equation (10). The mean and median error was
3.2 and 3.1%, respectively. With a 25th to 75th percentile range of 1.7–4.3%.
The difference in percent between c-factors based on volumetric density versus glandularity based on breast
density measurements was calculated. The mean and median error was 1.8 and 1.4%, respectively. The 25th
to 75th percentile range was 0.9 to 2.4%, and max error was 10.5%. The c-factor difference was largest for thin
breasts.

3.4.  Effect of measuring air kerma and beam quality

The difference between air kerma values based on measurements and the value reported in the DICOM
metadata, stratified by mammography unit and filter was calculated. The difference varies substantially
depending on mammography unit and filter. For the DM filters (Rh and Ag) the difference appears to be largely
constant with compressed breast thickness. For the DBT filter (Al) the difference varies by approximately 3–4%
with compressed breast thickness. The mean error in air kerma was 3.2%, and the mammography unit—filter
combination with the most error had a mean error of 7.6%.
The difference in between the HVL for each kilovolt peak (kVp) based on measurements and the value
reported in the DICOM metadata, stratified by mammography unit and filter was analyzed. For the DM filters
(Rh and Ag) the difference ranged from about 3 to 11%, increasing with kVp. The difference was smaller for the
DBT filter (Al), ranging from about  −1 to 6%, and was approximately constant with kVp.
The g-factor difference, depending on whether the g-factor was based on HVL measurements or reported by
DICOM metadata, was analyzed. The g-factor difference for the DBT filter (Al) ranged from about  −1 to 4%.
For the DM filters (Rh and Ag) the g-factor ranged from about 2 to 12%, and increased with compressed breast
thickness.

7
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Figure 2.  Variation in dose with thickness for Combo acquisitions of women with fatty (a) (quantized density 1 or 2) and dense (b)
(quantized density 3 or 4) breasts. And variation DBT/DM dose ratio as function of thickness for Combo acquisitions of women
with fatty (c) and dense (d) breasts. The colormap represents the number of views with resolution of 1 mm along the x-axis and 0.12
mGy (a) and (b) or 0.06 (c) and (d) along the y-axis.

The c-factor difference, depending on whether the c-factor was based on HVL measurements or reported by
DICOM metadata was analyzed. The median difference in c-factor was 0.2% ranging from  −1.4 to 0.3%.
Figure 4(a) shows the difference in AGD calculated using equation (1), depending on whether the model is
based on air kerma and HVL measurements or DICOM metadata. Here the glandularity of the breast is known
for both AGD calculations. The variation between mammography units and filters was substantial. The dose esti-
mates ranged from about  −5 to 12% and shows a clear trend with compressed breast thickness. The mean AGD
difference was 3.8%, but for one mammography unit as high as 7.9%.
Figure 4(b) shows the difference in AGD calculated by the Dance model based on measurements (glandular-
ity, air kerma and HVL) versus the AGD value reported by the DICOM header. Also here, there was substantial
difference between the different mammography units and filter combinations. However, the difference in dose
showed a larger variation. The mean and median AGD difference was  −10.7 and  −11.4%, respectively. For one
mammography unit, the AGD difference was as high as  −13.3%. The 25th to 75th percentile of  −14.2 to  −7.0%
with a range of  −21.7 to 18.4%.

4. Discussion

This study gives a detailed overview of the AGD in a paired DM and DBT setting, representative of mammography
screening of women in between 50 and 70 years. We also show that automatically calculated volumetric breast
density obtained using Quantra™ can be used to calculate glandularity consistent with the Dance dosimetry
model. The mean AGD was 1.74 and 2.10 mGy for DM and DBT, respectively. The mean DBT/DM AGD ratio was
1.24.
The European guidelines have acceptable and achievable AGD limits stratified by compressed breast thick-
ness (Perry et al 2006). Although the AGD values are meant to be measured in phantoms, the patient doses at
the phantom’s equivalent breast thickness are of interest. For the DM acquisitions in this study, the mean AGD
was within acceptable values for the thinnest beasts (up to 32 mm) and within achievable values for other breast
thicknesses. The acceptable values for DM are considered preliminary reference values for DBT acquisitions (van

8
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Figure 3.  (a) Shows the variation in glandularity calculated per the Dance model. A 3rd degree polynomial is fitted to the data
and compared to a fit by Dance et al (2000). (b) Shows the variation in c-factor with CBT for all acquisitions. (c) Shows the c-factor
difference due to using glandularities based on measured breast density and the fit to our data (equation (10)). The colormap
represents the number of views with resolution of 1 mm along the x-axis and 1% (a), 0.004 (b), and 0.5% (c) along the y-axis.

Engen et al 2016). Compared to DBT doses found in this study, the thinnest breasts (21 mm) are outside these
preliminary reference values, while other thicknesses are within. This suggests that further research is needed in
order to determine acceptable and achievable limits for DBT acquisitions.
The AEC of this type of mammography unit adjusts the exposure to account for the densest part of the breast
(Ren et al 2012) and is therefore sensitive to breast density. In this study, it was shown that the distribution of AGD
for DM is wider than for DBT. This indicates that the AEC adjusts more to differences in local densities in DM
mode than in DBT mode. Similar results have been shown by others (Bouwman et al 2015) who also reported
that the DBT mode is more sensitive to breast thickness, which is confirmed by data in this study. However, read-
ers should be aware that for most paired DM DBT acquisitions the HVL of the DBT acquisition is higher than for
the corresponding DM acquisition, especially for thicker breasts. Thus, the DBT spectra will be more penetrat-
ing. This will potentially also contribute to the reduction of the dose variation due to breast density for DBT-
compared to DM acquisitions.
For AGD stratified by breast density it was found that within each thickness strata for DM, dense breasts
received more dose than fatty, which was as expected as the AEC compensates for the density of the breast. How-
ever, for DBT the dose difference was minimal, which is consistent with the AEC primarily adjusting dose accord-
ing to breast thickness. The smaller dose difference is slightly positive for thinner breasts and slightly negative
for thicker breasts. The AGD ratio between DBT and DM shows that for the dense breasts the ratio is substanti­
ally lower for dense breasts than fatty, which also is consistent with the differences in AEC operation in the two
modes. Thus, the dose penalty of performing DBT on dense breasts was on average only 8% compared to 33% in
fatty breasts. For one unit (lab 2), there was about 70 cases for breast thicknesses of more than 55 mm resulting in
outliers (low dose) in DM acquisitions of primarily fatty breasts. These outliers reduce the DM dose substanti­
ally within the thickest strata (CBT 90–99 cm) in the fatty breasts columns in table 2 and in figure 1(a). These also
affect the DBT/DM dose ratio for the same strata. However, the effect of these few outliers in other strata and
overall results in this work is negligible.
Figure 3(a) shows the distribution of glandularity according to the glandularity model described by Dance.
The figure visually shows reasonably good agreement with glandularity calculated based on volumetric density

9
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

exported by Volpara™, shown in Tromans et al (2014) and Highnam (2017). As the cohort in this study (50–70
years) is older than the cohort used for Dance’s polynomial (50–64 years), the estimated glandularity is expected
to be lower, due to decrease in glandularity with age. This result indicates that it is feasible to calculate glandular-
ity based on the volumetric density values reported by Quantra™.
Comparing the volumetric density to glandularity, the volumetric density is lower, as expected. Thus, if volu-
metric density rather than glandularity is used in dose estimates, the dose estimate will be higher. This difference
is larger for thinner breasts as the 5 mm region of subcutaneous fat will occupy a larger proportion of the breast
for thinner breasts, resulting in a larger glandularity.
In equation (1), the c-factor is set to 1 if the glandularity is 50%. In this cohort, the average glandularity was
lower, resulting in an average c-factor of 1.16. Thus, ignoring the breast density, the breast dose will be underes-
timated by about 16%. Figure 3(b) shows that the c-factor is lower for thinner breasts, which is consistent with
monte carlo simulations by Dance et al (2000), as expected. The c-factor at constant CBT varies, as the c-factor
not only is dependent on CBT, but also the HVL and the glandularity if the breast.
A common way to estimate glandularity from breast thickness is to use the 3rd degree polynomial published
by Dance et al for one of two age groups (Dance et al 2000) shown in figure 3(a). Using this estimate for the most
appropriate age group reduces the underestimation of AGD to about 6%. This estimate was based on a younger
cohort than in this study, thus a higher c-factor was expected. In this work, a new 3rd degree polynomial (equa-
tion (10)) based on modern breast density assessment software was provided. This polynomial might be better
suited for a cohort between 50 and 70 years. Using this polynomial, over half of the women in the cohort would
have less than 3.1% error in their dose estimate due to error in glandularity. However, readers should be aware
that using a polynomial fit to obtain glandularity only provides a crude estimate on an individual patient basis, as
is reflected in the goodness of fit coefficient.
A common way to do dose assessment today is to use software for automatic data collection and dose calcul­
ation using DICOM metadata, such as VolparaDose™, DoseWatch™, DoseTrack™, tqm|dose™, Radimetrics™ or
custom-made applications. Most of these applications are either based on using DICOM metadata as input to
models like the Dance model, or use the AGD value reported directly. Figure 4(a) illustrates the use of dose calcul­
ation software with access to breast glandularity, but not calibrated air kerma and HVL data (similar to Volpara-
Dose™). Although a recent version of VolparaDose™ has the option of including calibrated air kerma and HVL
data. Figure 4(b) illustrates the use of dose collection software harvesting AGD from DICOM metadata. This
study shows that there can be substantial differences in air kerma and HVL depending on whether measurements
or DICOM metadata is used. These differences can vary substantially between mammography units, even if they
are the same model at the same site. Thus, measurements of both air kerma and HVL, can substantially reduce
the error in dose estimation. For the units at our site, the error in air kerma seemed to be constant with breast
thickness for DM, and slightly curved for DBT. Errors in HVL will also affect the dose estimates through errors
in the g- and c-factors. This study shows that it is primarily the g-factor which will affect the AGD due to errors in
HVL. For HVL, the errors in the Rh beam qualities seemed to increase linearly with kVp. For Ag and Al filter the
error was approximately constant. These results indicate that errors in air kerma and HVL can be corrected to
within reasonable accuracy, with measurements at only a few beam qualities at each unit. The largest AGD differ-
ence between two units was close to 10%. Such deviations would be of interest when comparing mammography
units within centers or within a national screening program. A within unit variation of about 8% was observed;
where DM doses for thinner breasts are underestimated compared to thicker breasts, while the DBT doses seem
to be slightly overestimated for thinner breasts compared to thicker. This would especially affect the comparison
of DM and DBT at different breast thicknesses. As expected the AGD reported in the DICOM metadata shows
more variation. The dose is underestimated by a mean of about 11%. This is of interest when comparing AGD
estimates between studies using AGD reported from the DICOM header.
In 2011 Olgar et al (2012) estimated AGD for DM and DBT in 641 patients (2247 DM and 984 DBT images)
for a Hologic Selenia Dimensions system. They measured air kerma and HVL and used Quantra™ (likely an older
version without skin correction) to assess volumetric breast density. The patient population was between 27 and
91 years. (mean 62). Their estimate of AGD (using the Dance model (equation (1))) was about 12% higher (1.92
mGy) for DM and 22% lower (2.55 mGy) for DBT and their dose ratio was about 8% higher (1.33), compared to
the results in this study.
Prior to 2013 Cavagnetto et al (2013) estimated AGD for DM and DBT for 300 consecutive patients of 38 to 84
years. (mean 54.2). They estimated AGD using the Dance model (equation (1)) while measuring air kerma using
MOSFET dosimetry. Glandularity was estimated using Dance polynomials (Dance et al 2000). Their estimates of
AGD were 1.31, 2.56 and 3.87 mGy for DM, DBT and Combo acquisitions respectively. Their estimates were 24%
lower, 22% and 2% higher for DM, DBT and Combo respectively.
In 2012 Shin et al (2015) reported AGD to 149 women between 22 and 78 years. (mean 51.1). Their AGD
estimates were obtained directly from the DICOM metadata. Mean AGD was 1.63 and 1.74 mGy for DM and
DBT, respectively. This was 5 and 17% lower than our estimates. They noted that DBT doses of thicker and denser

10
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Figure 4.  Differences in dose estimates using the Dance model based on measurements versus DICOM metadata values for air
kerma and HVL, where glandularity is known for both estimates (a). The dose estimate using the Dance model based on measured
values (glandularity, air kerma and HVL) versus the AGD value reported in the DICOM metadata (b).

breasts yielded doses similar to or maybe even smaller than for DM. Our study confirms this finding up to about
70 mm thick breasts.
Prior to 2015 Bouwman et al (2015) estimated AGD based on 2500 images of a selected cohort of women
between 19 to 88 years. (mean 51). They used the Dance model (equation (1)), measured air kerma and HVL, and
used the polynomials by Dance et al (2000) to estimate glandularity. Their dose estimates vary from 1.18 to 4.17
mGy for DM for 20–29 mm and 80–90 mm breasts, respectively. The corresponding doses in this study were 1.01
to 2.53 mGy. As the DM dose tables (AEC calibration) for these units were updated in January 2011, it is likely our
study used a different DM AEC calibration. For DBT their dose estimates varied from 1.53 to 4.64 mGy compared
to 1.27 to 3.70 mGy in our study, which is much closer than for DM.
From 2015 to 2016 Castillo-Garcia et al (2017) estimated AGD of 561 women between 32 to 86 years. (mean
54). They used the Dance model (equation (1)), measured air kerma and HVL and used volumetric and areo-
metric density calculated by Quantra™ directly. Their DM AGD estimates for women above 50 years are slightly

11
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

higher than in our study up to about 50 mm. For thicker breasts their dose remains about constant, while our
estimates continue to increase. For DBT, the AGD have a similar increase as this study, however their estimates
are consistently higher. This can partly be explained by using glandularity estimates directly and underestimating
the glandularity according to the Dance model, thus choosing a higher c-factor.
Gennaro et al (2018) analyzed AGD for 1208 women participating in the STORM-2 population based screen-
ing trial, imaged between 2013 and 2015. Their estimates were performed using VolparaDose™. Thus, they are
performed using the Dance model (equation (1)) with knowledge of the glandularity. However, air kerma and
HVL was not measured. Their AGD estimates were 1.37 and 1.87 mGy for DM and DBT, respectively. This is 20
and 11% lower than the results in our study. Comparing CBT between the study of Gennaro et al (2018) and our
study, the median CBT appear to be in the high end of the 41–50 mm interval, while in our study the median CBT
was 54 mm. This indicates differences in the screened cohort, which also may explain the differences in mean
AGD. Also, the air kerma and beam quality was to the authors knowledge not measured, which can potentially
add to systematic differences between the dose estimates.
As these studies, except for the STORM-2 trial, report dose estimates of selected populations. Comparison of
mean doses should be interpreted considering the selected cohort of women. As the AEC responds differently to
breast density in DM and DBT, and breast density has been shown to decrease with age, this will affect the dose
estimates. Additionally, several of the studies show substantial differences in DM doses. As the dose table of the
units in this study was updated in early 2011, the reported doses in this study will therefore only be comparable
with units using this updated setting. Incorporation of glandularity into the dosimetry model will also have a
substantial impact. Thus, estimates using volumetric breast density will overestimate dose, as only the central
region of the breast contains the glandular tissue in the Dance model. Additionally, the Dance model defines
glandularity as the fibroglandular fraction with respect to weight rather than volume, which will contribute to
further underestimation using volumetric density. Finally, comparison of dose estimates must consider whether
the air kerma and HVL has been measured or if DICOM metadata has been used directly. Using DICOM meta-
data will increase the uncertainty of the estimated AGD.
This study has some limitations. When the breast is compressed there is always some part of the breast which
is not. In this part, we do not know the thickness, which will contribute to the uncertainty of the breast density
assessment using Quantra™, and thus also in estimating the c-factor. Removing the uncompressed breast edge
can raise the estimated glandularity by 3–8% (Highnam 2017). The Dance model assumes the breast is images
in a CC projection, while in this study we have calculated AGD for both CC and MLO projections. A study by
Sechopoulos et al (2007) showed that the AGD to the MLO views can be overestimated by 11%. This was attrib-
uted to the fact that in the CC view the breast is positioned in the central ray, which has a higher photon fluence.
The MLO view will be more affected by the heel effect and a slightly longer focus to breast distance, which will
reduce the AGD compared to the CC view. Additionally, the Dance model assumes a semi cylindrical breast with
the glandularity homogeneously distributed in a central core. This is not a fully realistic model of the breast
and will introduce uncertainty in the dose estimates. Simulations in voxel phantoms have shown that errors
can be as large as 48% when comparing different glandular distributions in the breast (Dance et al 2005). Thus,
doses calculated in this study should be applied on a population level and not on individual subjects, as the
errors can be substantial. Also, these errors would affect calculations for DM and DBT similarly, thus this effect
on relative AGD between the two modalities will be minimal. The issue regarding glandular distribution could
potentially be solved in the future for modalities providing 3D information, such as DBT, by having monte carlo
based dosimetry based on the patient’s images (Dance and Sechopoulos 2016). However, as this approach would
require significant computational resources, it remains to see whether this will be practical. When assessing the
difference in air kerma, HVL and reported dose from DICOM metadata and measured values, only three dif-
ferent units from one manufacturer in one center were included. Thus, the differences reported here might not
generalize to other centers or manufacturers, however they illustrate that deviations between reported values and
measured values can be substantial.

5. Conclusion

In this population based study, AGD from paired same-compression DM and DBT acquisitions was estimated.
The mean dose for DM, DBT and Combo was 1.72, 2.09 and 3.81 mGy, respectively, with an average dose ratio
of 1.24. Denser breasts generally had a lower DBT/DM AGD ratio than fatty breasts, thus the dose penalty of
using DBT on dense breasts is substantially lower. It was also found that incorporating breast density into the
dosimetry model may have a substantial impact on dose. Although, estimating glandularity using previously
published polynomials improved the estimates, there will still be a systematic error related to differences in the
study cohort and the cohort, which is the basis of the polynomials. Finally, dose estimates derived from DICOM
metadata data potentially involves substantial uncertainty of which investigators should be aware.

12
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Acknowledgments

We would like to thank Inger Eide Sterner for assistance in exporting a large number of examinations.

References
Becket J R and Kotre C J 2000 Dosimetric implications of age related glandular changes in screening mammography Phys. Med. Biol.
45 801–13
Bernardi D, Macaskill P, Pellegrini M, Valentini M, Fantò C, Ostillio L, Tuttobene P, Luparia A and Houssami N 2016 Breast cancer screening
with tomosynthesis (3D mammography) with acquired or synthetic 2D mammography compared with 2D mammography alone
(STORM-2): a population-based prospective study Lancet Oncol. 17 1105–13
Boone J M, Fewell T R and Jennings R J 1997 Molybdenum, rhodium, and tungsten anode spectral models using interpolating polynomials
with application to mammography Med. Phys. 24 1863–74
Bouwman R W, van Engen R E, Young K C, den Heeten G J, Broeders M J M, Schopphoven S, Jeukens C R L P N, Veldkamp W J H and
Dance D R 2015 Average glandular dose in digital mammography and digital breast tomosynthesis: comparison of phantom and
patient data Phys. Med. Biol. 60 7893–907
Carney P A et al 2003 Individual and combined effects of age, breast density, and hormone replacement therapy use on the accuracy of
screening mammography Ann. Int. Med. 138 168–75
Castillo-García M, Chevalier M, Garayoa J, Rodriguez-Ruiz A, García-Pinto D and Valverde J 2017 Automated breast density computation
in digital mammography and digital breast tomosynthesis Acad. Radiol. 24 802–10
Cavagnetto F, Taccini G, Rosasco R, Bampi R, Calabrese M and Tagliafico A 2013 ‘In vivo’ average glandular dose evaluation: one-to-one
comparison between digital breast tomosynthesis and full-field digital mammography Radiat. Prot. Dosim. 157 53–61
Dance D R 1990 Monte Carlo calculation of conversion factors for the estimation of mean glandular breast dose Phys. Med. Biol. 35 1211–9
Dance D R and Sechopoulos I 2016 Dosimetry in x-ray-based breast imaging Phys. Med. Biol. 61 R271–304
Dance D R, Hunt R A, Bakic P R, Maidment A D A, Sandborg M, Ullman G and Alm-Carlsson G 2005 Breast dosimetry using high-
resolution voxel phantoms Radiat. Prot. Dosim. 114 359–63
Dance D R, Skinner C L, Young K C, Barrett J F and Kotre C J 2000 Additional factors for the estimation of mean glandular breast dose using
the UK mammography dosimetry protocol Phys. Med. Biol. 45 3225–40
Dance D R, Young K C and van Engen R E 2009 Further factors for the estimation of mean glandular dose using the United Kingdom,
European and IAEA breast dosimetry protocols Phys. Med. Biol. 54 4361–72
Dance D R, Young K C and van Engen R E 2011 Estimation of mean glandular dose for breast tomosynthesis: factors for use with the UK,
European and IAEA breast dosimetry protocols Phys. Med. Biol. 56 453–71
Gennaro G, Bernardi D and Houssami N 2018 Radiation dose with digital breast tomosynthesis compared to digital mammography: per-
view analysis Eur. Radiol. 28 573–81
Hammerstein G R, Miller D W, White D R, Masterson M E, Woodard H Q and Laughlin J S 1979 Absorbed radiation dose in mammography
Radiology 130 485–91
Hartman K, Highnam R, Warren R and Jackson V 2008 Volumetric assessment of breast tissue composition from FFDM images Breast
Imaging vol 5116, ed E A Krupinski (Berlin: Springer) pp 33–9
Highnam R 2017 Patient-Specific Radiation Dose Estimation in Breast Cancer Screening Keeping Patients Safe and Informed (Wellington:
Volpara Solutions) (www.volparasolutions.com/assets/Uploads/VolparaDose-White-Paper.pdf) (Accessed: 9 May 2017)
Hologic 2012 Understanding Quantra™ 2.0 User Manual Man-02004 Rev 004 (Bedford: Hologic) (www.hologic.com/sites/default/files/
package%20inserts/Understanding%20Quantra%202.0%20User%20Manual.%20English.pdf) (Accessed: 9 May 2017)
IAEA 2007 Dosimetry in diagnostic radiology: an international code of practice Technical reports series no. 457 IAEA, Vienna (www-pub.
iaea.org/MTCD/publications/PDF/TRS457_web.pdf)
IAEA 2011 Quality Assurance Programme for Digital Mammography: IAEA Human Health Series No. 17 (Vienna: IAEA) (www-pub.iaea.org/
MTCD/publications/PDF/Pub1482_web.pdf)
Johns P C and Yaffe M J 1987 X-ray characterisation of normal and neoplastic breast tissues Phys. Med. Biol. 32 675–95
McCormack V A and dos Santos Silva I 2006 Breast density and parenchymal patterns as markers of breast cancer risk: a meta-analysis
Cancer Epidemiol. Biomarkers Prev. 15 1159–69
Olgar T, Kahn T and Gosch D 2012 Average glandular dose in digital mammography and breast tomosynthesis RoFo 184 911–8
Perry N, Broeders M, de Wolf C, Törnberg S, Holland R and von Karsa L (ed) 2006 European Guidelines for Quality Assurance in Breast
Cancer Screening and Diagnosis 4th edn (Luxembourg: European Commision, Office for Official Publications of the European Union)
Perry N, Broeders M, de Wolf C, Törnberg S, Holland R and von Karsa L (ed) 2013 European Guidelines for Quality Assurance in Breast
Cancer Screening and Diagnosis 4th edn (Luxembourg: European Commision, Office for Official Publications of the European
Union) pp 14–20 (www.euref.org/downloads?download=46:supplement-4th-edition-european-guidelines) (Accessed: 9 May 2017)
(Supplements)
Pisano E D et al 2005 Diagnostic performance of digital versus film mammography for breast-cancer screening N. Engl. J. Med. 353 1773–83
Rafferty E A, Durand M A, Conant E F, Copit D S, Friedewald S M, Plecha D M and Miller D P 2016 Breast cancer screening using
tomosynthesis and digital mammography in dense and nondense breasts JAMA 315 1784–6
Ren B, Smith A P and Jing Z 2012 Local versus whole breast volumetric breast density assessments and implications Breast Imaging vol 7361,
ed A D Maidment et al (Berlin: Springer) pp 775–85
Sechopoulos I, Suryanarayanan S, Vedantham S, D’Orsi C J and Karellas A 2007 Computation of the glandular radiation dose in digital
tomosynthesis of the breast Med. Phys. 34 221–32
Shin S U, Chang J M, Bae M S, Lee S H, Cho N, Seo M, Kim W H and Moon W K 2015 Comparative evaluation of average glandular dose and
breast cancer detection between single-view digital breast tomosynthesis (DBT) plus single-view digital mammography (DM) and
two-view DM: correlation with breast thickness and density Eur. Radiol. 25 1–8
Skaane P et al 2013 Prospective trial comparing full-field digital mammography (FFDM) versus combined FFDM and tomosynthesis in a
population-based screening programme using independent double reading with arbitration Eur. Radiol. 23 2061–71
Skaane P, Bandos A I, Eben E B, Jebsen I N, Krager M, Haakenaasen U, Ekseth U, Izadi M, Hofvind S and Gullien R 2014 Two-view digital
breast tomosynthesis screening with synthetically reconstructed projection images: comparison with digital breast tomosynthesis
with full-field digital mammographic images Radiology 271 655–63

13
Phys. Med. Biol. 63 (2018) 035006 (14pp) B H Østerås et al

Svahn T M, Houssami N, Sechopoulos I and Mattsson S 2015 Review of radiation dose estimates in digital breast tomosynthesis relative to
those in two-view full-field digital mammography Breast 24 93–9
Tagliafico A S, Tagliafico G, Cavagnetto F, Calabrese M and Houssami N 2013 Estimation of percentage breast tissue density: comparison
between digital mammography (2D full field digital mammography) and digital breast tomosynthesis according to different BI-
RADS categories Br. J. Radiol. 86 20130255
Tromans C E, Highnam R, Morrish O, Black R, Tucker L, Gilbert F and Brady M 2014 Patient specific dose calculation using volumetric
breast density for mammography and tomosynthesis Breast Imaging vol 8539, ed H Hujita et al (Berlin: Springer) pp 158–65
van Engen R E et al 2016 Protocol for the Quality Control of the Physical and Technical Aspects of Digital Breast Tomosynthesis Systems. Version
1.01 (Nijmegen: EUREF) (www.euref.org/downloads?download=53:european-tomo-qc-protocol-version-1-01) (Accessed: 9 May
2017)
Yaffe M J 2008 Mammographic density. Measurement of mammographic density Breast Cancer Res. 10 209–19
Yaffe M J and Mainprize J G 2011 Risk of radiation-induced breast cancer from mammographic screening Radiology 258 98–105
Young K C, Ramsdale M L and Rust A 1996 Dose and image quality in mammography with an automatic beam quality system Br. J. Radiol.
69 555–62

14