1.

Purpose

Even though digital breast tomosynthesis (DBT) is a well-established imaging modality, efforts to optimize system accuracy and efficiency are still ongoing. Currently, different modes of gantry and/or detector movement exist. The two main acquisition modes for acquiring DBT projection data are a continuous gantry motion mode and a step-and-shoot mode. In the continuous mode, the x-ray tube moves continuously while x-rays are generated in short pulses at specific angles around the breast. This has the potential for faster acquisition times compared with the step-and-shoot operation but incurs source motion blurring along the direction of the source motion. Source motion blurring has been shown to reduce technical image quality, as quantified by a number of metrics.^1–5 In a step-and-shoot mode, the x-ray tube stops at each angular location to acquire a projection image. This eliminates source motion blur provided that the tube stops without inducing any mechanical instability but may result in a longer scan acquisition time and therefore an increased chance for patient motion artifacts.⁶

It is likely that a wide-angle DBT system using a step-and-shoot approach to eliminate source motion blur would result in unacceptably long acquisition times. This work investigates the technical performance of a novel x-ray tube technology with a “flying focal spot” (FFS). Although the x-ray tube with the anode target still moves continuously during the DBT scan, the electron beam moves in the opposite direction compared with the tube movement during the x-ray pulse. The term “flying” therefore refers to the opposite movement of the electron beam, which in fact results in a focus that remains stationary for each projection. This technology is expected to reduce or eliminate source motion blurring while allowing for reduced scan acquisition time.⁷

This study compares two wide-angle DBT systems with two different acquisition modes: the continuous mode and the new FFS technology with a stationary focus during image acquisition. The impact on image sharpness and visibility of calcification- and mass-like test objects is quantified using technical image quality metrics.

2.

Methods

Two DBT systems were investigated in this study: a Siemens MAMMOMAT Revelation and a Siemens MAMMOMAT B.brilliant. Both systems acquire 25 projections over an angular range of 50 deg. The Siemens MAMMOMAT Revelation employs the continuous motion acquisition mode, which incurs blurring due to focus motion during each projection.^8,9 Hereafter, this is referred to as the “uncompensated focus” system. On the other hand, the Siemens MAMMOMAT B.brilliant is equipped with an x-ray tube with an FFS to mitigate source motion blurring in the acquired projection images. The FFS is utilized exclusively during acquisitions involving gantry movement, where compensation for source motion is needed. Therefore, in the digital mammography (DM) mode, both systems operate with a static focus, i.e., the FFS mechanism is not active when the gantry is not rotating. The technical specifications of both systems can be found in Table 1. In addition, the B.brilliant system incorporates a next-generation DBT reconstruction algorithm.⁷ The algorithms of both systems are based on a filtered back projection algorithm with non-linear iterative optimization steps.

Table 1

System specification comparison of two wide-angle DBT systems: one with an x-ray tube with moving focus during image acquisition (Siemens MAMMOMAT Revelation) and one with an x-ray tube with FFS technology (Siemens MAMMOMAT B.brilliant).

3.

Results

3.1.

X-ray Tube and Characterization of Geometric Blurring

The focus size results are given as (length in the tube-travel direction) × (length in the front-back direction). In the DM mode, the focus sizes for the Revelation were $(0.39\pm 0.01)\times (0.50\pm 0.01)\mathrm{mm}$, whereas focus sizes for the B.brilliant were $(0.27\pm 0.01)\times (0.56\pm 0.01)\mathrm{mm}$. Similar focus size results were observed in the DBT mode for the Revelation, with values of $(0.33\pm 0.01)\times (0.47\pm 0.01)\mathrm{mm}$. Slightly larger dimensions were measured for the B.brilliant in DBT mode: $(0.41\pm 0.02)\times (0.79\pm 0.02)\mathrm{mm}$. The DBT focus size measurements do not include source motion blur as they were obtained in 0-deg angle stationary mode. For both systems, the largest focus dimension was in the front-back direction, as can also be seen in the focus images in Fig. 1. Note that some degree of dose dependency is expected for the measured focus size.

Fig. 1

DM and DBT “For Processing” image of the focal spot of a MAMMOMAT Revelation and a MAMMOMAT B.brilliant, acquired with a multi-pinhole test object. DBT images were acquired in 0-deg angle stationary mode. The horizontal axis is the front-back direction, and the vertical axis is the tube-travel direction.

Figure 2(a) shows that the DM MTF curves are similar for both systems. There is very little influence of focus size on the presampling MTF in DM mode as the edge was positioned on the breast support table, minimizing geometric blurring. Figure 2(b) shows the MTF curves measured in DBT projections in the front-back direction. The MTF gradually decreases with increasing height above the table due to focus size geometric blurring. The MTF for the B.brilliant system with the FFS is slightly lower compared with the Revelation system, consistent with the larger focus observed for this system, especially in the front-back direction. In the tube-travel direction, there is an additional blurring due to source motion for the Revelation, which results in a significant decrease in MTF as a function of height [Fig. 2(c)]. This may also be present for the B.brilliant with FFS if the x-ray tube motion is not entirely canceled by the electromagnetic deflection of the focus; however, this was not seen in these measurements. Figure 3 shows an image of the edge in the central DBT projection for the two systems, along with a graph of the ESF in which some broadening of the edge transition can be seen for the Revelation compared with the B.brilliant.

Fig. 2

Average MTF in front-back and tube-travel direction obtained from five DICOM “For Processing” DM images (a) and five central DBT projection images (b), (c) acquired on a MAMMOMAT Revelation and a MAMMOMAT B.brilliant. The dash-dotted curves were obtained at 2-cm height in 0-deg angle stationary DBT mode, excluding source motion blur. MTF values at 4 and 6 cm in 0-deg mode, not shown, were similar to those at 2 cm.

Fig. 3

Central DBT projection of MTF edge in the tube-travel direction, together with their ESF measured at 2, 4, and 6 cm above the table using a MAMMOMAT Revelation and a MAMMOMAT B.brilliant.

For B.brilliant with the FFS in DBT mode, $f_{25\%}$ at 2, 4, and 6 cm above the table was respectively $7.1\pm 0.2$, $6.7\pm 0.1$, and $6.0\pm 0.1{\mathrm{mm}}^{-1}$ (front-back) and $7.6\pm 0.2$, $6.8\pm 0.1$, and $6.6\pm 0.1{\mathrm{mm}}^{-1}$ (tube-travel). The difference in $f_{25\%}$ in the front-back and tube-travel direction is smaller for B.brilliant compared with Revelation, resulting in an improved MTF isotropy for B.brilliant. The FFS tube in the B.brilliant gives notably higher $f_{25\%}$ values in the tube-travel direction compared with the Revelation system. The $f_{25\%}$ values at 2, 4, and 6 cm above the table were respectively 35%, 49%, and 59% higher.

The MTF in DBT reconstructed planes, i.e., the in-plane MTF, gives information on the blurring present in the DBT projection images but also includes the influence of image reconstruction on image sharpness. A different reconstruction algorithm is implemented on both systems, with each algorithm being optimized for the acquisition geometries of its respective system. The in-plane MTF measured in the tube-travel direction at different heights in the reconstructed DBT planes and normalized to the curve peak is shown in Fig. 4. The corresponding LSFs are shown in Fig. 5. The in-plane MTF in the tube-travel direction deteriorates with increasing height. The degradation is most pronounced for the Revelation system, which shows a 41% reduction in $f_{25\%}$ from 2 to 6 cm, compared with a 12% reduction for the B.brilliant system. The drop in MTF in the low spatial frequency region ($<1{\mathrm{mm}}^{-1}$) for both systems is probably due to the application of a ramp filter in the reconstruction.

Fig. 4

In-plane MTF measured in the tube-travel direction at different heights in reconstructed DBT slices of a MAMMOMAT Revelation and a MAMMOMAT B.brilliant.

Fig. 5

In-focus reconstructed DBT slice of tungsten wire running perpendicular to the chest wall edge imaged on top of 2, 4, and 6 cm PMMA, together with the LSF. Both LSFs have some overshoot in the tails, which is more pronounced in the B.brilliant data, causing a reduction in MTF at spatial frequencies below $\sim 1{\mathrm{mm}}^{-1}$.

Figure 6 shows the spatial frequency where the MTF drops to 25% ($f_{25\%}$) measured in DBT projections as a function of time for the B.brilliant system with FFS. The deviations are within 10% of the average $f_{25\%}$ value over a month of measurements. This demonstrates that the FFS technology is stable over time.

Fig. 6

Spatial frequency of the 25% MTF value in the front-back (a) and tube-travel (b) direction measured in DBT projections of the MAMMOMAT B.brilliant over one month.

A DBT acquisition on the Revelation system requires $\sim 21.7\mathrm{s}$, whereas this is reduced to 4.8 s for the B.brilliant system. This results in a gantry speed of $2.3\mathrm{deg}/\mathrm{s}$ for the Revelation compared with $10.4\mathrm{deg}/\mathrm{s}$ for the B.brilliant. If the gantry speed of the Revelation would be increased to $10.4\mathrm{deg}/\mathrm{s}$, without compensating for source motion, then the focus size due to motion would rise from 1.9 to 8.8 mm. To evaluate the impact of focus motion, the DBT projection MTF was modeled as the product of the presampling MTF and a sinc function describing the projected focus length due to motion.^8,9 As a result, $f_{25\%}$ for the DBT projection MTF in the tube-travel direction at 40 mm above the breast support table would theoretically decrease from 4.3 to $1.1{\mathrm{mm}}^{-1}$ (Fig. 7), resulting in unacceptable blurring of the projections. The x-ray tube filtration has been changed from Rh to Al in the B.brilliant, which will allow higher tube outputs, enabling an increased acquisition rate for the same total scan dose.¹⁷ This significant reduction in acquisition time has the potential to reduce patient motion artifacts.

Fig. 7

Analytically modeled DBT projection MTF in the tube-travel direction at 4 cm above the breast support table for the B.brilliant system with clinical rotation time (4.8 s), the Revelation system with clinical rotation time (21.7 s), and the Revelation system if it would operate with 4.8-s rotation time.

3.2.

CDMAM as a Function of Height

Figure 8 shows the threshold gold thickness for the 0.16-mm diameter discs of the CDMAM phantom positioned at different heights above the table. The results of the 0.16-mm discs are presented to avoid any extrapolation for the reported threshold gold thicknesses. Nonetheless, similar results were observed for smaller-diameter CDMAM discs.¹⁸

Fig. 8

Threshold gold thickness for the 0.16 mm discs retrieved from the CDMAM phantom imaged in DM (a) and DBT (b) mode on a MAMMOMAT Revelation and MAMMOMAT B.brilliant. Threshold gold thicknesses were corrected for MGD. Error bars indicate the 95% confidence interval.

From system design parameters, we might expect some differences in the DM results between the systems. Although both systems use the same model of x-ray detector and anti-scatter grid, and a static focus, the B.brilliant has an Al filter compared with the Rh filter in the Revelation. The initial contrasts of the gold discs in the phantom are higher for the Rh-filtered spectrum, which might give some improvement in threshold gold thickness performance. However, in the DM mode, similar threshold values are seen for the two devices, once adjusted for dose. For both systems, the detection threshold increases with height above the table as a result of geometric blurring by the extended focal spot.

In the DBT mode, the detection thresholds improved for the system with FFS, particularly at greater heights above the table. The $T_t$ values at 25, 45, and 65 mm above the table improved by 12%, 16%, and 24%, respectively, for the B.brilliant compared with the Revelation. These results are generally consistent with the measured in-plane MTF curves. However, the CDMAM results will also be affected by the anode/filter combination, as in the DM mode, and by the reconstruction algorithm. Figure 9 shows the in-focus reconstructed DBT slice of the CDMAM phantom at 65 mm above the table for both systems.

Fig. 9

In-focus reconstructed DBT slice of the CDMAM phantom positioned 65 mm above the table, acquired on a MAMMOMAT Revelation (a) and a MAMMOMAT B.brilliant (b). The images display variations in window-leveling due to different x-ray spectra and reconstruction algorithms used by the systems.

3.3.

L1 Phantom as a Function of Height

The reader-averaged percentage correct (PC) values of each lesion in the L1 phantom are presented in Fig. 10 for all three image types, for the two systems. A substantial inter-reader agreement was found (ICC = 0.77). In DM, no performance difference in detecting calcifications within the L1 structured background was observed between the two systems ($p=0.73$), irrespective of lesion height. These results are consistent with the CDMAM results, even though detection was here studied in processed DM images. The B.brilliant system demonstrated a significantly improved detection of non-spiculated masses ($p=0.012$), but only a slight, non-significant improvement for spiculated masses.

Fig. 10

Reader-averaged percentage of correctly detected targets (PC) in L1 images acquired on a MAMMOMAT B.brilliant against images acquired on a MAMMOMAT Revelation. Error bars represent the standard error of the mean.

Overall, the B.brilliant showed better detection of all three lesion types in DBT images compared with the Revelation ($p<0.0001$). The odds of detecting calcifications were 1.56 times higher with B.brilliant compared with Revelation (95% CI [1.21, 2.02]; $p=0.001$). For the 180 to $200\mu \mathrm{m}$ calcification cluster located 54 mm above the table, the PC dropped by almost 30% for Revelation compared with a PC of 97% for B.brilliant [see arrow Fig. 10(b)]. This difference can also be observed in Fig. 11. However, no significant impact of calcification height was observed ($p=0.37$) contrary to what was found with CDMAM.

Fig. 11

Visualization of a calcification cluster in the L1 phantom with three different modalities using a MAMMOMAT Revelation and a MAMMOMAT B.brilliant. The cluster consists of five calcifications, each 180 to $200\mu \mathrm{m}$ in diameter, arranged in a dice pattern and positioned at heights of 24 and 54 mm above the table.

Some improvement was also seen in the SM images of B.brilliant compared with Revelation, showing significantly improved detection of calcifications and spiculated masses. The odds of detecting a calcification cluster in the L1 phantom were 1.80 times higher for B.brilliant compared with Revelation (95% CI [1.42, 2.27]; $p<0.0001$). For example, the PC of the 180- to $200\text{-}\mu \mathrm{m}$ calcification cluster at 24- and 54-mm height was 43% and 31% higher, respectively, for the B.brilliant system [see arrows Fig. 10(c)]. Again, no significant effect of calcification height was observed ($p=0.07$). When comparing the three imaging modalities, the odds ratio for lesion detection remains higher in both DM (Revelation: OR = 1.72, 95% CI [1.49, 1.98]; B.brilliant: OR = 1.20, 95% CI [1.03, 1.39]) and DBT (Revelation: OR = 3.47, 95% CI [2.98, 4.04]; B.brilliant: OR = 3.48, 95% CI [2.96, 4.10]) compared with SM ($p<0.0001$).

4.

Discussion

The FFS technology in the Siemens MAMMOMAT B.brilliant system results in an improved MTF isotropy in DBT projections compared with the Siemens MAMMOMAT Revelation, which suffers from source motion blur. The MTF isotropy is now comparable to DBT systems that use the step-and-shoot approach.¹⁹ This suggests that the implementation of the flying focus adequately compensates for x-ray tube motion during the acquisition of each projection image, as the gantry moves around the breast in a DBT scan. Moreover, it has been demonstrated that this compensation remains consistent over a period of one month.

The FFS tube has a slightly larger focus area in the DBT mode, and this had some impact on the MTF in the front-back direction measured in the projections for B.brilliant versus Revelation. Nevertheless, the B.brilliant offers superior resolution in the tube-travel direction compared with Revelation. Zheng et al.⁴ demonstrated that a typical focal spot size up to $\sim 0.3\mathrm{mm}$ does not impact the reconstructed image resolution if the x-ray tube remains stationary during data acquisition. Conversely, source motion blur significantly degrades the reconstructed image resolution as also observed in this study.

Although there is a definite improvement in the in-plane MTF curve at 6 cm above the table for B.brilliant in the tube-travel direction, in-plane MTF results at 2 and 4 cm are similar for the two systems. The increase in the projection MTF for the B.brilliant is not seen in the in-plane MTF results for these heights, and this may be due to differences in processing. The MTF measured in-plane clearly includes the influence of spatial frequency filtering performed in the reconstruction algorithm.²⁰ It is possible that the high in-plane MTF seen for the Revelation at 2 and 4 cm comes at the expense of amplified noise and the reduction seen at 6 cm comes from the strong geometric blurring at this height.

There are some limitations to the measurement of in-plane MTF using either a wire, edge, or disc^9,21,22 and further depend on the test object used for the measurement. The presence of additional processing methods such as artifact reduction, which are object dependent, can also have a strong influence. Nevertheless, previous work²² on various DBT systems using different (including non-linear) reconstruction algorithms showed a reasonably good correlation between the detectability of small diameter discs and a Fourier-based detectability metric. In trying to explain the detectability of small details such as microcalcifications, the MTF cannot be used in isolation, and aspects such as signal profile and noise must also be considered.²³ However, reducing geometric blurring in the acquired projection image dataset gives greater flexibility in the successive reconstruction and image processing steps for the system with the FFS tube.

Threshold gold thickness measured using the CDMAM test object in the DM images was lower for the Revelation system at 15 and 25 mm above the table, whereas slightly superior performance was seen for the B.brilliant at 65 mm. Overall, the FFS tube had no net impact on the technical image quality of the DM images. Spatial resolution, as quantified using MTF, and the detection of calcification- and mass-like objects in a structured background were similar in the DM mode for the two systems, despite differences in x-ray spectrum and resulting initial contrast, as well as differences in the image processing algorithms.

An improved small detail detectability in DBT mode was observed for the B.brilliant system. For the CDMAM phantom used in this study, notable improvements in the threshold gold thickness for 0.16-mm-diameter discs as well as smaller discs (not reported) were found, especially when the phantom was positioned at 55 mm or higher above the table. In addition to the increased, isotropic resolution, this may also be partly attributed to the reconstruction algorithm, which is known to influence lesion visibility.²⁴ It should be noted that object detectability quantified using contrast-detail test objects depends on the signal-to-noise ratio (and therefore the dose level)²⁵ and not just the resolution or sharpness of the detector and imaging system.²⁶

The aim of this work was to compare the imaging performance with an FFS tube against that of an x-ray tube with an uncompensated focus. However, given the difference in anode/filter settings and processing algorithms, a like-for-like comparison was not possible. Manufacturer-determined AEC settings were used for imaging the test objects, and CDMAM results were corrected for differences in MGD to account for variations in quantum noise in the images. Nevertheless, some differences remain due to the difference in the x-ray spectrum.

In DM mode, the systems are set up to meet the image quality and dose requirements specified in the European DM protocol.²⁷ After MGD correction, the CDMAM results in Fig. 8 show that only small differences exist between the two systems. For B.brilliant, the tube motion compensation is not activated in the DM mode, and therefore, any differences arising from the x-ray spectrum, scattered radiation, anti-scatter grid and x-ray detector are likely to be small.

In the DBT mode, the reconstruction algorithm will influence the threshold gold thickness results. The anode/filter combination may also slightly influence the results, although we would expect this effect to be similar or even smaller due to the thinner Al filter used in DBT compared with the DM mode in the B.brilliant. The MGD-corrected threshold gold thickness results are significantly lower for the B.brilliant in the DBT mode, consistent with the elimination of geometric blurring due to a compensated net focus motion relative to the breast during exposure.

The CDMAM phantom is designed to evaluate image quality in 2D mammography and must be used with caution when comparing performance among different DBT systems.^28,29 The CDMAM results only provide information on the in-plane detectability of small details, without any influence from background structures and do not demonstrate the absolute performance of both DBT systems. Both systems have an angular range of 50 deg, and therefore, the influence of the tomography angle on the magnitude of in-plane structured noise will be very similar. There may however be differences due to the reconstruction algorithm, which can influence the artifact spread function and therefore the level of in-plane anatomical noise.¹⁹

PMMA was used to support the CDMAM phantom when assessing threshold detectability at various heights above the table. Although this allowed the use of clinically relevant technique factors for the different breast equivalent thicknesses, the use of PMMA increased the amount of scattered radiation present, especially for DBT acquisitions as there is no anti-scatter grid present in this mode. Scatter will reduce threshold detectability; however, the effect is expected to be similar for both systems. Both units use the same detector technology and the same anti-scatter grid in DM mode and no grid in DBT mode. The difference in x-ray beam energy between the systems only has a small influence on the scatter-to-primary ratio,³⁰ and therefore, the impact of scatter on small detail detectability is likely comparable for both systems.

To support the CDMAM results, calcification detection was also assessed using a 3D phantom with a structured background. Improved lesion detection in DBT was again observed for the B.brilliant, although the effect of the height of the lesion above the breast support table was less pronounced than in the CDMAM results. One possible reason for this is the greater uncertainty associated with these results as they were obtained with human readers rather than automatic readout as used for CDMAM.¹⁵ The background structure clearly plays a role in the greater uncertainty of the measured detection thresholds. In addition, it was observed that the current calcification size groups in L1 were either visible at a 100% PC level or not visible at all; there were almost no calcification diameters with a 50% to 80% PC detection rate that exhibited subtle visibility. Smaller effects that impact detection, such as the height of the lesion, may not be detected with the current set of calcification diameters in the L1 phantom.

When lifting the L1 phantom above the table, we chose not to use PMMA supports, unlike the CDMAM phantom. Combining 30 mm of PMMA with the L1 phantom would result in a breast-equivalent thickness of $\sim 100\mathrm{mm}$,¹¹ which is not very common in clinical practice. Furthermore, using PMMA would alter the L1 phantom composition, change the structured noise generated by the phantom,¹⁵ and potentially change the difficulty of the detection task. For these reasons, we opted to use 30 mm of polystyrene instead. This choice ensures that lesion detectability results between the two phantom thicknesses are not influenced by changes in background structure and are therefore more likely to be related to differences in system performance parameters. The primary goal of this study is to compare the two systems using like-for-like phantom compositions whenever possible.

There was a difference in MGD between the systems when imaging the L1 phantom, which was difficult to correct because the effect of dose on the signal-to-noise ratio within a structured background does not follow the Rose model as is the case for the homogeneous CDMAM phantom.³¹ The AEC configuration of the B.brilliant gave a 13% higher MGD for the L1 phantom in DM compared with the Revelation and a 7% higher MGD for DBT. Based on the work of Vancoillie et al.,³¹ the impact of this small dose difference on detectability was considered negligible when comparing both systems.

The SM images of the L1 phantom demonstrated that the improved sharpness and isotropy in DBT mode for the B.brilliant system were exploited in the SM processing, resulting in the improved detection of calcifications and spiculated masses compared with the Revelation system. Nevertheless, the performance of SM has not yet reached the performance achieved in DM and DBT as seen in technical image quality evaluations for other DBT devices.^31,32

The FFS x-ray tube in the B.brilliant system reduced the DBT scan acquisition time, from 21.7 s for the Revelation to 4.8 s for a 45-mm breast equivalent thickness. If this scan time would be implemented in the Revelation system without FFS technology, we could expect a drop of $\sim 70\%$ in $f_{25\%}$ for the DBT projection MTF at 40 mm above the breast support table.²⁰ Ghani et al.³ observed a decrease of $\sim 17\%$ in cutoff frequency (10% MTF) of the in-plane MTF when doubling the gantry speed in continuous motion mode for a narrow-angle DBT system. The FFS technology allows wide-angle DBT scans to be performed with a scan time similar to that found in narrow-angle systems.¹⁹ This has clear potential to reduce patient motion artifacts, which will likely improve image quality.³³ An additional benefit of the continuous gantry motion used with the FFS tube compared with a step-and-shoot method might include the elimination of small vibrations in the gantry and breast support platform as the gantry accelerates and decelerates during the scan.

5.

Conclusion

The introduction of an FFS x-ray tube in the Siemens MAMMOMAT B.brilliant system brings established technology from CT imaging to DBT imaging systems. This has greatly improved MTF isotropy in DBT projections, effectively eliminating the relative motion of the source during exposure. As a result, scan speed has been increased by a factor of $\sim 4$ without introducing source motion blur, an issue that would severely degrade image sharpness if the same gantry rotation speed was applied to the Siemens MAMMOMAT Revelation. Compared with the Revelation, which has a moving (uncompensated) focus during exposure, the B.brilliant system with FFS tube demonstrated superior resolution in the DBT mode and improved small detail detectability, especially for details located at greater heights above the table, quantified using a contrast-detail method. Furthermore, mass and calcification detectability within a 3D structured background improved compared with the previous generation of tube technology. Together with the reduced DBT acquisition time, FFS technology mitigates the limitations associated with source motion, leading to improved small detail detectability.