Table of Contents
Lag and Ghosting
Over recent years, a number of studies have been published that demonstrated the advantages of CMOS-based X-ray detectors for Digital Breast Tomosynthesis (DBT)1,2. When compared to thin-film transistor (TFT) passive pixel sensor technology, CMOS active pixel sensors provide several advantages such as faster frame rate, low electronic noise, and smaller pixel pitch3. While the electronic noise of the CMOS sensor described in this article is below 200 e- 4, TFT detectors for mammography generally show a considerably higher electronic noise, which could reach 1000 e- and even higher. The reduced pixel size for TFT technology will come at the cost of fill factor, whereas the 49.5 μm CMOS sensor described in this article has a fill factor of 79%. Low noise levels and small pixel pitch are required for detecting small microcalcifications5,6. To detect early-stage breast cancer, the detection of microcalcifications smaller than 200 μm is very important. Park et al.7 demonstrated that calcifications could be resolved using an X-ray detector with a ~75 μm pixel 165 μm. With a pixel of around 50 μm, it is possible to detect microcalcification of ~100 μm.
Full Field Digital Mammography (FFDM) needs higher saturation dose levels than DBT mode. A high dynamic range is required to integrate both modes into a single detector. This can be realized by designing two full well capacities into the CMOS sensor to obtain two saturation dose levels8,9, one full well capacity optimized for FFDM and the other for DBT applications. This article presents a different approach known as frame-summing. Several image frames are acquired and summed in a single X-ray pulse. The benefit of this approach is that the saturation dose level for FFDM mode can be adjusted through the number of frames summed.
The 23x29 cm2 mammography detector contains four wafer-scale CMOS sensors having a small 49.5x49.5 μm2 pixel as well as an on-chip 14-bit analog-to-digital converter with low power consumption and low readout noise. The mammography detector functions at 8 frames per second in full resolution and 16 fps in 2x2 pixel binning mode. X-rays are converted to visible light by a CsI scintillator customized for mammography applications.
The FFDM application typically sets the saturation dose requirement of a mammography detector. If the detector should also serve the DBT application, then a different dose range (usually 10% of the FFDM dose) needs to be mapped onto the full ADC range. This is usually achieved by applying a gain switch. As a result, only a fraction of the available signal storage capacity of the pixels is used and this is accompanied by reduced signal-to-noise performance. This way of integrating DBT and FFDM in a single detector is schematically shown in Figure 1. Given the low dose levels in DBT, the reduction in signal-to-noise ratio is a major drawback of gain switching.
Here, a method is presented that resolves this drawback and provides excellent signal-to-noise performance for both DBT and FFDM in a single detector. The detector presented in this article has a pixel saturation dose of roughly 1 mGy and is improved for the more complicated DBT application. In FFDM mode, the detector is read out a number of times during the X-ray exposure and the frames are summed to acquire the final FFDM image. Owing to the high frame rate achievable with the CMOS technology, the pixels will not saturate and the performance signal-to-noise ratio improves because the summing operation averages the noise and boosts the output bit depth. Figure 2 shows a schematic illustration of this frame summing technology.
Figure 1. FFDM & DBT modes using gain switch.
Figure 2. FFDM & DBT modes using frame summing.
In order to maintain excellent X-ray image quality by this implementation, several requirements have to be met. The frame rate together with the saturation level must be able to accommodate the maximum application dose rates. The detector is developed for a maximum 10 mGy/s dose rate in full resolution mode, which is more than enough for FFDM applications. The detector response vs. dose should be highly linear to prevent artifacts. During frame-summing operation, it is not allowed to lose any X-ray dose. To achieve this, the CMOS detector is operated in continuous rolling shutter mode such that the photodiode is invariably integrating. The detector’s noise level needs to be low and the frame rate setting has to be improved for the dose rate so that saturation in any part of the image can be prevented by every single frame.
The linearity of the mammography detector in FFDM mode, and in DBT mode is shown in Figure 3. The measurements were made at RQA-M2 radiation quality (Mo/Mo, 28 kV, 2 mm Al). In DBT mode, the X-ray pulse was exposed inside a single frame capture. The detector’s integration time was 923 ms and the maximum output level was 16384 DN, which elucidates the deviation of the last measurement point in the graph.
In FFDM mode, four frames are obtained in sequence and then summed within the detector, with the output being the single summed image. The integration time was 250 ms per frame, leading to a frame rate of 4 fps. The maximum output level of ~56000 DN is slightly less than 4x16384 DN, due to the continuous rolling shutter readout mode. In this rolling shutter readout mode, the detector is read out in a row by row sequence while all rows excluding the one read out are continuously integrating. The readout time of one row is extremely short 42 μs and does not lead to any meaningful loss in X-ray dose.
The readout of one complete frame takes approximately 123 ms. To capture an X-ray pulse, the pulse has to end before the start of the last frame readout. This reduces the integration time of the last frame by 123 ms. The number of frames summed decides the saturation dose in FFDM mode and this could be adjusted to certain application needs. A sequence of any number of frames can be captured and summed.
Excellent linearity is shown by both FFDM mode and DBT mode. The achieved linearity graph is not affected by summing frames. The sensitivity in both FFDM mode and DBT mode is also identical demonstrating that there is no loss in X-ray dose. At low exposure conditions, minor inaccuracies in dose measurements will lead to larger deviations from linearity in percentage terms.
Figure 3. Blue curve shows the sensitivity in DBT (left) and FFDM (right) mode measured at RQA-M2. Green curve shows the deviation from a linear fit through the measurement points.
The MTF evaluation of the detector is shown in Figure 4. Measurements were made in accordance with the IEC 62220-1-2 standard at RQA-M2, 280 μGy. Although the performance of the MTF is lower than advanced a-Se direct conversion detectors (at 5 lp/mm MTF=35% vs MTF=50% for a-Se10), it is better than other CMOS-based mammography detectors (MTF at 5 lp/mm = 30%)1,2
The DQE performance in FFDM mode and in DBT mode is shown in Figure 5. The DQE was measured in accordance with the IEC 62220-1-2 standard at RQA-M2. For FFDM and DBT modes, the same detector settings as for the linearity measurements were used. For DBT mode, the DQE was measured at two dose levels, a low dose of 27 μGy and a normal dose of 285 μGy at the detector entrance surface. The DQE in FFDM mode was measured at 2 mGy and 285 μGy at the detector entrance surface.
Figure 4. The MTF performance at RQA-M2, 285 uGy.
The DQE shows similar performance for FFDM and DBT modes at the same X-ray exposure of 285 μGy. For FFDM, 4 frames were summed in the detector while for DBT mode, the DQE was achieved with a single image capture. At high dose levels, above saturation level in DBT mode, the DQE performance in FFDM mode remains the same at 285 μGy and shows that the performance is not affected by summing.
This graph demonstrates that the detector provides good image quality in both DBT and FFDM modes and that it exceeds competitive detectors particularly at higher spatial frequencies1,2,10. In addition, the imaging experiments on the CDMAM phantom confirm the excellent DQE performance in both FFDM and DBT modes.
DQE measurements using RQA-M2 beam quality at very low dose conditions were unavailable and therefore a model11 was used to demonstrate the low noise performance of the CMOS detectors. It has been demonstrated that this modeling can well predict the low dose performance of CMOS-based X-ray detectors for different noise and saturation dose levels as well as different optical stack configurations9. The quantified low dose performance (green curves) at 10 μGy and 5 μGy are plotted based on the measurements at 285 μGy, as shown in Figure 6. The calculated and measured performance at 27 μGy is shown for comparison purposes. The DQE performance begins to decrease below 20 μGy, but at 5 μGy the DQE is still larger than 35% at 5 lp/mm.
Figure 5. DQE performance at RQA-M2 in DBT mode (single image capture) and FFDM mode (4 frames summed) at typical dose conditions.
Figure 6. The measured (blue) and modeled (green) DQE performance in DBT mode at RQA-M2.
Figure 7. The measured DQE performance as a function of dose in DBT mode at 28 kV, W/Rh.
In Figure 7, a different mammography spectrum is used to plot the DQE as a function of dose. For this measurement, a W/Rh anode/filter combination was used at 28 kV with another 2 mm Al filter. Higher DQE values are obtained for this spectrum, and the DQE does not start to decrease until 10 μGy.
Lag and Ghosting
Lag and ghosting are key parameters for good image quality in DBT mode. Lag is the residual signal in a dark image produced by previous X-ray exposures and was determined measured with the detector operated in rolling shutter mode at 10.5 fps. A W-anode, 40 kV, and 2 mm Al filter were used to perform the measurements. Following an X-ray pulse of 3 mGy with a duration of 2 seconds, the remaining signal was measured in dark, that is, without any X-ray exposure. The result is plotted in Figure 8. The lag is already smaller than 0.17% after 0.1 seconds. This is considerably better than reported for a-Se detectors12 and shows a valuable benefit of CMOS detector technology.
Ghosting refers to the change in X-ray sensitivity due to previous radiation exposure. To measure ghosting, half of the detector is first shielded with lead and then a 15 mGy exposure is applied to the detector. By definition, ghosting is the sensitivity difference between the irradiated regions relative to the regions covered with lead. The ghosting result is plotted in Figure 9. The measurement reveals that the ghosting is 0.08% after 5 minutes and 0.16% after 90 seconds. The CsI scintillator dominates the lag and ghosting performances.
Figure 8. Lag after 3 mGy X-ray exposure during 2s.
Figure 9. Ghosting after covering half of the detector with lead and irradiating with15 mGy X-ray exposure.
This article has presented the performance and characteristics of a high dynamic range CMOS X-ray mammography detector developed to support both FFDM and DBT applications, demonstrating that it indeed provides excellent image quality for both applications and favorably compares with advanced a-Se mammography detectors. At higher spatial frequencies, the CMOS X-ray mammography detector outperforms in DQE performance, presenting the opportunity to detect microcalcifications of ~100 μm. The implementation of frame summing provides a high dynamic range. This feature also provides the possibility to implement cutting-edge patient dose management functionality during the procedure.
For instance, the X-ray pulse duration required to attain a certain dose in a particular region of the image can be determined from the first frame of the image, so that the dose given to the patient can be optimized for the clinical imaging task.
The authors gratefully acknowledge Bart Dillen for all his valuable discussions on the implementation of frame-summing into the detector and Yves Kessener for carefully proofreading the manuscript. This work was supported by the European CATRENE program, CAT406-NEMADE.
 Zhao, C., Kanicki, J., Konstantinidis, A.C. and Patel, T., “Large area CMOS active pixel sensor x-ray imager for digital breast tomosynthesis: Analysis, modeling, and characterization,” Med. Phys. 42 (11), 6294-6308 (2015)
 Zhao, C., Konstantinidis, A.C., Zheng, Y., Anaxagoras, T., “50 μm pixel pitch wafer-scale CMOS active pixel sensor x-ray detector for digital breast tomosynthesis,” Phys. Med. Biol. 60 (2015), 8977–9001 (2015)
 Zentai, G., “Comparison of CMOS and a-Si flat panel imagers for X-ray imaging,” Proc. IEEE IST 05/211, 194-200 (2011)
 Weisfield, R.L., Bennett, N.R., “Electronic noise analysis of a 127 μm pixel TFT/photodiode array,” Proc. SPIE 4320, 209-218 (2001)
 Burgers, A.E., “Effect of Detector Element Size on Signal Detectability in Digital Mammography,” Proc. SPIE 5745, 232-242 (2005)
 Wheeler, F.W., Perera A.G.A., Claus, B.E., Muller S.L., Peters G., Kaufhold, J.P., “Micro-calcification detection in digital tomosynthesis mammography,” Proc. SPIE 6144, (2006)
 Park, H.S., Kim, Y.S., Kim, H.J., Choi Y.W., Choi J.G., “Optimization of configuration parameters in a newly developed digital breast tomosythesis system,” J. of Radiat. Res 55, 589-599 (2013)
 Konstantinidis, A.C., Szafraniec, M.B., Speller, R.D., Olivo A., “The Dexela 2923 CMOS X-ray detector: A flat panel detector based on CMOS active pixel sensors for medical imaging applications”, Nuclear Instruments and Methods in Physics Research A 689, 12–21 (2012)
 Maes, W.H., Peters I.M., Smit, C., Kessener, Y., Bosiers, J., “Low-dose performance of wafer scale CMOS-based X-ray detector,” Proc. SPIE 9412, (2015)
 Granfors, P.R., Aufrichtig R., Possin, G.E., Giambatista B.W., Huang Z.s., Jianqiang, L., Bing, M., “Performance of a 41x41 cm2 amorphous silicon flat panel x-ray detector designed for angiographic and R&F imaging applications,” Med. Phys. 30 (10), 2715-2725 (2003)
 Zhao B., Zhao W., “Imaging performance of an amorphous selenium digital mammography detector in a breast tomosynthesis system,” Med. Phys. 35 (5), 1978-1987 (2008)
This information has been sourced, reviewed and adapted from materials provided by Teledyne DALSA.
For more information on this source, please visit Teledyne DALSA.