Key words
image quality - detective quantum efficiency - digital breast tomosynthesis - modulation
transfer function - noise power spectrum
Introduction
Digital breast tomosynthesis (DBT) is a 3 D imaging technique that can reduce or eliminate
the tissue overlap effect. During tomosynthesis, the X-ray gantry rotates and acquires
projections of the stationary compressed breast at multiple angles [1]. The projection images are then reconstructed into a series of thin-slice images.
Breast tomosynthesis systems have been developed by different manufacturers, and an
increasing number of studies compare the clinical performance of FFDM and DBT [2]
[3]. There are also some recent publications dealing with the dosimetric properties
of DBT for the 2 D conventional mammography and 3 D breast tomosynthesis modes [4]
[5].
Image quality can be assessed subjectively or quantitatively [6]
[7]
[8]. Subjective image quality assessment is observer-dependent, and the observer can
be adapted to the simulated objects in the phantoms used in the measurements. Therefore,
quantitative image quality measurements with established standards give more realistic
results than visual methods.
Quantitative image quality of the digital system can be measured in terms of detector
response or signal transfer property (STP), modulation transfer function (MTF), noise
power spectrum (NPS) and detective quantum efficiency (DQE). The MTF is used to describe
the resolution properties and the NPS is used to characterize the magnitude of the
noise of an imaging system in the spatial frequency domain. A higher MTF means better
image sharpness and resolution, and the lower the NPS, the lower the noise within
of radiographic image. The DQE is always derived from MTF and NPS measurements to
quantify the overall image quality and compare the performance of different image
detectors quantitatively. The DQE is defined as the efficiency of the imaging detector
in transferring the signal-to-noise ratio (SNR) from an input to an output. The detectability
of low-contrast objects is highly dependent on the imaging system’s SNR and the magnitude
of noise. To increase detectability of low-contrast lesions, the SNR should be increased,
whereas image noise should be reduced as much as possible, as an ideal detector DQE
is equal to unity at all spatial frequencies [9]. In practice, the SNR at the output of the imaging system is always lower than the
input, and the DQE value of an imaging system is always less than unity. The higher
the DQE, the more X-ray photons interacting with the imaging detector are used to
produce an image.
The IEC62 220 – 1-2 standard was introduced in 2007 to standardize imaging geometry
and beam quality in DQE measurements for digital mammography systems [10]. The implementation of this standard permits the comparison of the DQE among different
imaging detectors. Several studies have reported image quality measurement results
for different breast tomosynthesis systems [11]
[12]
[13]
[14]
[15]. Most of these studies were carried out using a Siemens Mammomat prototype tomosynthesis
system (Siemens, Erlangen, DE). To our knowledge, the first physical image quality
measurements for the Selenia Dimensions breast tomosynthesis system were carried out
by Hologic employees [16]
[17]. Marshall and Bosmans recently published a study of system sharpness measurement
(MTF) comparing Siemens Inspiration and Hologic Selenia Dimensions DBT systems [18].
Our study characterizes the MTF, NPS and DQE of an amorphous selenium-based Hologic
Selenia Dimensions breast tomosynthesis system using established methods. The DQE
was measured for different detector air kerma values and compared with published results
from other breast tomosynthesis systems.
Materials and methods
The Selenia Dimensions (Hologic, USA) digital mammography system can be used both
for 2 D conventional mammography and 3 D breast tomosynthesis [16]. The Selenia Dimensions image detector is based on direct capture amorphous selenium
technology and has a detector pixel size of 70 µm. The detector is operated in full
resolution mode with a pixel size of 70 µm in 2 D mode and 2 × 2 binning with a pixel
size of 140 µm in 3 D mode. In 2 D imaging mode, an anti-scatter high transmission
cellular (HTC) grid which automatically moves out of the field of view when 3 D imaging
mode is used. The X-ray tube has a tungsten (W) anode with additional filtration of
50 µm rhodium (Rh), 50 µm silver (Ag) and 0.7 mm aluminum (Al). The Rh or Ag X-ray
filter is used in the 2 D imaging mode, and the Al filter in the 3 D imaging mode.
During breast tomosynthesis the system acquires 15 projection images in increments
of approximately 1° starting at –7.5° and ending with + 7.5°, with the breast in standard
compression. The focus detector distance is 70 cm, and the breast support plate is
2.5 cm above the detector surface. The acquired image of the breast at different angles
is reconstructed by using a specialized filtered back projection method [16]
[17].
The first step of the quantitative image quality measurements is the determination
of the detector response, which gives the relationship between the mean pixel value
(PV) and the detector air kerma. The detector air kerma was measured with a calibrated
dosimeter (UNIDOS webline, PTW Freiburg, Germany) and a mammographic ion chamber (SFD
Chamber Type 34 069, PTW Freiburg, Germany). The accuracy of dose measurements was
5 %. The dose was measured as a function of mAs with 28 kVp, W/Rh target filter in
2 D standard projection mammography mode and W/Al target filter setting in 3 D breast
tomosynthesis mode. Detector response measurement was carried out with flat field
zero degree tomo mode, and the mean pixel value was taken from the first image of
the tomography sequence to reduce the lag effect in 3 D breast tomosynthesis mode.
In this mode, the X-ray tube was stationary during image acquisition. Individual DBT
projection images in 3 D breast tomosynthesis mode were extracted from archive files
using “gexpand” and “mview” decoding software provided by Hologic Corp. The detector
response was measured following the geometry described in IEC protocol for mammography
and high purity (99.9 %) 2 mm Al added filtration to the beam [10]. The mean pixel value was extracted from a region of interest (ROI) of 256 × 256
pixels placed at the distance of 60 mm from the chest wall edge. The PV was then plotted
against the detector air kerma (DAK). The detector response curve is used to normalize
and linearize the images used for the calculation of MTF and NPS.
The standard deviation (SD) of the PV within the ROI was also recorded during detector
response measurement in order to investigate quantum limited operation of the detector.
The SNR was calculated from the measured mean pixel value and the SD of the PV within
the ROI, and the SNR2 was plotted against the detector air kerma. The linearity of this curve was established
by plotting a best fit through all measured points [19].
The presampling MTF is measured using the edge method as described elsewhere [20]. The edge test device consists of a 1 mm thick, 120 mm long and 60 mm wide stainless
steel plate. The edge was placed on the breast support plate with the edges angled
approximately 30 to the pixel matrix during measurement. The edge spread function (ESF) of the system
was defined as the pixel value versus the perpendicular distance from the edge transition.
The ESF was then differentiated to obtain the line spread function (LSF). Finally,
the MTF was calculated via Fourier transformation of the LSF. The images for MTF calculation
were acquired at 28 kVp using a W/Rh target/filter combination and without additional
filtration or grid in the 2 D standard projection mammography mode. The same exposure
condition was used in 3 D mode but with a W/Al target filter combination. The MTF
was measured for both c-arm scanning (tube-travel direction) and front-back (chest
wall-nipple direction) directions. In the results section, only the MTF in the scanning
direction is reported because of small differences between the MTF results and to
show the overall system performance.
The NPS was calculated from the detector response images using previously published
methods [21]
[22]
[23]. Sub-images of 1024 × 1024 pixels were extracted from the central region of the
flat field images and linearized to air kerma using the response curve. A 2 D second-order
polynomial surface was fitted and then subtracted from this region in order to remove
low frequency background trends such as the anode-heel effect from the X-ray source
on the NPS. Half overlapped (by 64 pixels in each direction), regions of interest
(ROIs) of 128 × 128 pixels were selected from the sub-image. In this manner, a total
of 128 ROIs were used in the NPS measurements. The NPS was calculated by implementation
of 2 D fast Fourier transform (FFT) to each ROI using software developed by NHSBSP
[24]. One dimensional (1 D) NPS was obtained from 2 D NPS by averaging central ± 7 rows
or columns (including the axis) around each axis. The normalized noise power spectrum
(NNPS) was then calculated by dividing by the square of the mean PV of the linearized
sub-image. DQE can be calculated from the measured MTF and NNPS as follows:
where MTF(u) is the measured pre-sampling MTF in the u direction (the tube scanning direction
in this study), DAK is the measured air kerma at the detector plane for flat field image acquisition,
NNPS(u) is the normalized noise power spectrum in the u direction and q is the number of photons per unit air kerma per mm2 for the X-ray beam quality used in the NNPS and detector response measurements [25]. The factor q is calculated from the software MIQuaELa v.1.0 package [26]
[27].
Discussion
In this study, the image quality of the Hologic Selenia Dimensions breast tomosynthesis
system was measured quantitatively. Quantitative image quality measurements have been
published for various breast tomosynthesis systems. However, only a few studies focused
on the breast tomosynthesis system used in this study.
The detector response curves established in our study were linear and similar to the
curves reported by Marshall et al. for the Siemens Mammomat InspirationTomo [11]. Many digital imaging systems demonstrate linear response curves (e. g., flat panel
X-ray imaging systems), while some have logarithmic response curves (e. g. photostimulable
phosphor detector systems). A detector having logarithmic response produces a signal
more directly related to the tissue composition along the path, due to the exponential
attenuation of X-rays. Logarithmic transformation can be applied to the detector data
with the effect of reducing the range of the signal [28]. However, in order to make meaningful calculations, the relationship between system
input and output should be linear. If the STP is logarithmic or a power law, all the
image data should be linearized by applying the inverse of the STP to each pixel value
[29]. Marshall et al. reported that the gradient of the detector response for the DBT
mode was higher than the 2 D imaging mode by a factor of 3.5. In this study, we measured
a detector response gradient in DBT mode 3.1 times greater than that in 2 D mode because
of the 3 D mode operated with 2 × 2 pixel binning and with a higher electronic gain.
This allows a fast readout, suppression of other electronic noise after pre-amplification
and the usage of a more dynamic range of the detector, resulting in signal saturation
at lower detector air kerma values [16].
Our MTF measurement results for different imaging modes were also similar to Marshall
and Bosmans’s results for the same breast tomosynthesis system used in this study
[18]. They reported MTFs for 2.0 and 4.0 spatial frequencies of approximately 0.78 and
0.60 for the planar mammography (2 D), 0.60 and 0.22 for DBT scan mode, and 0.58 and
0.12 for the reconstructed planes in left-right (LR) direction, respectively. The
measured MTFs for the same spatial frequencies in our study were 0.81 and 0.59 for
2 D, 0.66 and 0.34 for DBT scan, and 0.58 and 0.13 for the reconstructed plane in
the LR direction, respectively. Our MTFs were also comparable to results from Baorui
et al. for the prototype Hologic breast tomosynthesis system [17].
The measured NNPS in this study was comparable to the NNPS measured by Marshall et
al. for Siemens Mammomat InspirationTomo [11]. Their NNPS value at 2.0 spatial frequency and 23.9 µGy DAK value is approximately
1.0 × 10-5 in DBT scan mode and 3.0 × 10-6 at 98.0 µGy DAK value in 2 D mode. In this study, we calculated an NNPS of 8.32 × 10-6 at 22.4 µGy DAK value and 2.12 × 10-6 at 103.5 µGy DAK value in DBT scan tomo mode and 2 D mode, respectively.
In this study, the calculated DQE was saturated at approximately 13.5 µGy in 3 D scan
tomo mode and between 47.3 and 74.0 µGy in 2 D planar mammography mode. The DQE was
saturated at a lower DAK value in 3 D imaging mode due to the higher gradient of the
detector in this mode. The established maximum DQE value was 54 % for both 2 D and
3 D imaging modes. The maximum DQE value was decreased for higher DAK values at lower
frequencies in the 2 D imaging mode. The gain map acquired at lower detector pixel
counts failed at higher counts. After gain correction of the high count images, more
artifacts could be detected. At lower frequencies, these artifacts increased and the
DQE dropped. For normal 2 D breast imaging, the detector rarely measures above 1000
counts. The average level is about 500 counts in 2 D imaging mode. The flat field
gain map used for correction at this level results in the best image quality. A correct
DQE measurement at higher count levels should be performed with a new gain map.
Baorui et al. reported higher DQE values for the same breast tomosynthesis system
used in this study [16]. They measured the DQE without a detector cover, and the difference in DQE values
can be attributed to the detector cover used during DQE measurement in our study.
The DQE value saturated at lower DAK values (9.43 µGy in 3 D and between 48.8 and
73.3 µGy in 2 D mode) in the study by Baorui et al. because the calculated DAK value
at the detector surface in our study is approximately 10 % higher. We measured air
kerma values used in DQE calculation above the breast support plate, which resulted
in an overestimation of the air kerma value at the detector entrance. This situation
in turn will lead to an underestimation of the calculated DQE for a specific air kerma
value by the reciprocal of the transmission of the breast support table in terms of
air kerma [22]. Choi et al. measured the physical image quality and evaluated the clinical performance
of the prototype DBT system developed by KERI (Korea Electrotechnology Research Institute)
and reported a maximum DQE for the full resolution and 2 × 2 pixel binning modes as
47.09 % and 50.24 %, respectively [30]
[31].
Zaho et al. reported closer DQE values for the prototype of Siemens Mammomat NovationDR. They established DQE values for both full resolution (pixel size 85 µm) and binning
mode (pixel size 170 µm in the tube travel direction) at DAK values between 3. 6 – 50
µGy. They described a maximum DQE value around 55 %, which is in good agreement with
our findings [13]. Bissonnette et al. have also reported comparable DQE values for the Siemens Mammomat
NovationDR at 2. 3 – 43.6 µGy detector air kerma range [15]. Varjonen estimated the DQE for a prototype GE Diamond DX breast tomosynthesis system
in the 7 – 136 µGy DAK range. She measured the MTF with a slit phantom and established
the DQE using a 28 kVp, Mo/Mo target/filter combination and 4 cm PMMA included in
the beam path. Her result for the maximum DQE value is higher than our results [14]. This difference may have resulted from the beam quality and the different MTF measurement
method used in her study.
Conclusion
This paper reports detailed measurements of quantitative image quality of a Hologic
Selenia Dimensions breast tomosynthesis system. The measured DQE values for different
detector air kermas were comparable with breast tomosynthesis systems from Siemens
and GE.
