Utilization of an energy-resolving detection system for mammography applications: A preliminary study

The ACR mammography phantom is designed to assess the image quality of mammographic systems. The phantom itself contains 16 test objects: six fibers, five microcalcifications clusters, and five masses. The masses have different volumes with a diameter of 12, 9, 7, 5, and 4 mm. For a mammographic system to pass the mammography quality standard act ‘MQSA’, at least four fibers, three microcalcification clusters, and three masses must be visible [20]. This quality assessment criterion is used to visually evaluate the quality of the images acquired through this study.

The experimental setup consisted of an Am-241 source, a scanning system, a lead collimator, NaI (Tl) detector, and the ACR mammography phantom. Am-241 is an alpha emitter with a half-life of 432.6 years and emits mainly 59.5 keV (35.9%) and 26.3 (2.3%) [21]. The Am-241 used in this study is housed in a shutter system and has an activity of 1 Ci. The source is cylindrically shaped and has a radius of about 1.5 cm. The experimental setup is shown in Fig. 1.

Fig. 1

The ACR phantom is placed in front of the source and held by a mechanical arm. The mechanical arm is controlled by the scanning system. The scanning system specifies the arm movement and the duration of the scan. The arm can be moved vertically or horizontally or in a raster scan.

The collimator is placed on top of the detector to preserve resolution and define the pixel size. It is 2 mm thick and has a 1 mm opening. Smaller openings are possible to fabricate but much more difficult to properly align with the rest of the setup. Additionally, smaller size openings reduce the count rate and therefore increase scanning time. These obstacles are somewhat easier to overcome in simulation; therefore, smaller openings were used in the simulation.

In addition to the collimator, the detector is enclosed within a lead sleeve to prevent scattered photons from reaching the detector’s active volume, causing an increase in the dead time.

When the source is turned on, the arm moves the phantom into the first acquisition position. Photons that penetrated the phantom reach the detector collimator. The collimator is aligned with the beam central axis. It blocks radiation outside the field of view from reaching the detector. Detected photons spectroscopic information is processed by the scanning system. After the defined acquisition time elapses, the scanning system stores the spectroscopic data as a function of the acquisition position and moves into the next position. Once the scan is complete, the system uses the acquisition position and spectroscopic information to construct an image, either using the entire spectrum or specific energy channels.

In the simulation, an 11 cm × 11 cm × 4.4 cm phantom with similar characteristics and properties to the ACR mammography phantom was created and the same assessment criteria were used to evaluate the simulated images using Geant4 Application for Tomographic Emission (GATE). GATE is one of the prominent Monte Carlo simulation toolkits with a wide range of applications in the field of medical imaging [22]. The simulation setup consisted of an Am-241 point source, two lead collimators, a NaI detector, and a phantom similar to the ACR mammography phantom as shown in Fig. 2. The number of primers we generated for each energy (59.5 keV and 26.3 keV) was 8 million photons which correspond to the activity of 348 MBq of Am-241. The standard physics processes were enabled in the simulation that included photoelectric effect, Compton scattering, and electron ionization.

Fig. 2

The source was placed at a 20 cm distance from the phantom. The emitted radiation is directed toward the phantom as in a cone beam configuration. The beam passes through a 4 mm thick pre-phantom and post-phantom collimators to limit the beam to the scanned area. The collimator square openings used in the simulation are 0.5 mm and 1 mm, respectively. The post-phantom collimator is placed on top of the NaI detector. The source, detector, and collimators remain stationary all the time while the phantom moves only between acquisitions. The collimator opening size dictates the phantom movement. For a 1 mm opening, the phantom would take a 1-mm horizontal or vertical step between acquisitions. The number of counts for each acquisition is recorded for the 26.3 keV and 59.5 keV windows separately. The acquired data are later used for constructing 2D images using MATLAB.

The source was initially placed at 10 cm from the detector to maximize exposure and reduce the scanning time. Images acquired under this setup exhibited high counting losses due to dead time and pile-up events. As a result, none of the phantom test objects were visible for both energy windows reconstructed images.

To find the optimum distance, two areas in the phantom were imaged at different distances. The first area enclosed the center of the largest tumour. The second enclosed the phantom center (no test object is present in that area). It was evident that the difference in counts between the two points is more distinct at a 35 cm distance. At that distance, scanning the whole phantom would require several hours. Therefore, greater distances were not attempted as it would increase the scanning time further. Very long scanning times caused the system to become irresponsive. Therefore, each phantom object was scanned separately. The acquisition time was set to 150 s/pixel. Reconstructed images for both the 26.3 keV and 59.5 keV windows are illustrated in Fig. 3 for the region containing the largest tumour.

Fig. 3

The tumour is visible with both energy windows. However, it is more evident (shows better contrast) with the 26.3 keV window, as the attenuation difference between the tumour and normal breast tissue is higher at 26.3 keV. Additionally, 26.3 keV photons have a higher absorption probability in the phantom. As a result, fewer photons reach the detector, and therefore, 59.5 photons produce a higher signal-to-noise ratio image. Figure 4 shows the energy spectrum of a single tumour pixel from both the 26.3 keV and the 59.5 keV.

Fig. 4

Typically, the maximum allowed kVp for mammographic X-ray tubes is 40. Higher kVps are not used as the attenuation difference between normal and cancerous cells becomes insignificant beyond 40 keV. Additionally, at higher energies, image contrast deteriorates due to scattered photons becoming the dominant interaction. By eliminating the scattered photons signal component from the image, contrast is expected to improve, and the mammographic energy range could possibly be extended to higher energies.

For comparison, several images were acquired using imaging detectors widely used clinically, such as computed radiography (CR) and flat panel. These detectors are pixelated and therefore, the setup was static during acquisition. The exposure time was set initially to 150 s and was increased gradually up to two days. Additionally, a 1-mm collimator was attached to the source opening to reduced geometrical blurring. However, none of the ACR mammography phantom test objects were detected in all images obtained with the flat-panel and CR detectors. Figure 5 shows a sample image obtained with a 70-micron pixel size computed radiography detector and a 2-day exposure. The incapability of commonly used imaging detectors to detect the largest tumour, even with much higher resolution than our scanning system, distinguishes energy-resolving detectors over other currently clinically employed imaging detectors. Some of the test objects, such as the microcalcifications, have higher attenuation differences than tumours with normal tissue. However, they are much smaller in size. Therefore, they are expected to be detected with an energy-resolving detector with a resolution higher than the one employed in this study.

Fig. 5

The undetectability of test objects at higher energies shows a distinctive advantage of the energy-resolving detector over other commonly used imaging detectors. Considering the relatively poor resolution of the NaI scanning system, even with relatively poor resolution, it was capable of detecting one of the test objects.

The ideal setup alignment and the absence of signal losses resulted in simulated images having better quality than the experimental ones. Figure 6 illustrates images produced with a 1 mm step. At least three tumours were visible with 26.3 keV photons and two tumours with 59.5 keV photons.

Fig. 6

The 59.5 keV images were produced by irradiating the phantom with 22 MBq of Am-241 (8 MBq of 59.5 keV photons). 26.3 keV photons images, on the other hand, were produced with 348 MBq of Am-241, where 8 MBq of these photons are 26.3 keV. Increasing the activity reduces the image noise but does not increase the number of visible objects. Improving the resolution by using 0.5 mm steps increases the number of visible objects for both energy windows, as illustrated in Fig. 7. Fibers show similar behaviour to the tumours as they have similar attenuation coefficients. Three out of six fibers are detected with both energy windows, with the fourth fiber being partially detected. The fourth fiber is very narrow in size and most of its pixels are shared with the background. As a result, its signal is reduced by the averaging effect.

Fig. 7

Tumours are easier to detect, with the used imaging setup, as they occupy larger areas. On the other hand, microcalcifications are the smallest in size; therefore, more difficult to detect. A single group of microcalcification can be partially detected for both energy windows with 0.5 mm openings, as illustrated in Fig. 8.

Fig. 8

The detected cluster only occupies four pixels with each pixel embodying a single microcalcification. Although microcalcifications are the smallest in size, they have the largest attenuation difference with the phantom background. This means that they are typically easier to detect if the imaging system has enough resolution; therefore, more microcalcification can be detected if smaller steps were used.